No entries match these filters. Clear filters
Xalgorix now keeps single-host scans from touching third-party domains found in the target's pages, redirects, or JavaScript︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.192 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.192
Xalgorix now adds a replayable curl command to every HTTP/HTTPS HTTPS HTTP carried over an encrypted TLS connection. It protects traffic in transit and lets a browser verify the server's certificate. finding that lacked one, so every proof of concept can be reproduced in one line.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.190 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.190
Every HTTP/HTTPS finding now includes a copy-pasteable curl command in its proof of concept, built from the finding's reported method and endpoint when the agent did not already produce one — literal ../ path-traversal segments, URL encoding, and shell-quoting are preserved so the request replays exactly.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
1 detail in Everything
Xalgorix adds wider SSRF SSRF Server-side request forgery, an attack that tricks a server into making requests chosen by an attacker. Those requests can reach internal services the attacker cannot access directly. -safe host parsing and expands file-tool credential-file protections︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.189 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.189
└──▷ TRY IT
Inspect the current deny-list to confirm new credential-file entries are active after upgrading.
copy
$ curl -s -H 'Authorization: Bearer <token>' http://127.0.0.1:9137/api/status | jq '.read_deny'
str_replace_editor, search_files, and list_files now refuse xalgorix.env files, the provider key store and auth profile store in Data_Dir and their temporary copies, and /proc/*/environ.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet The deny-list now blocks file-tool writes in addition to reads, so a path in XALGORIX_READ_DENY_LIST blocks both operations. Not checked yet The read_deny list in /api/status now shows the newly added deny-list entries. Not checked yet search_files with ripgrep now searches the path it was given and includes hidden and ignored files, matching grep -r behavior. Not checked yet
str_replace_editor, search_files, and list_files now refuse xalgorix.env files, the provider key store and auth profile store in Data_Dir and their temporary copies, and /proc/*/environ.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
3 more in Everything
└──▷ BREAKING ON UPGRADE
!
Entries added through `XALGORIX_READ_DENY_LIST` now also block file-tool writes — any path previously allowed for writing via the write allow-list that also appears in `XALGORIX_READ_DENY_LIST` will be refused on write.
Our best guess
xalgorix adds SSRF SSRF Server-side request forgery, an attack that tricks a server into making requests chosen by an attacker. Those requests can reach internal services the attacker cannot access directly. -safety checks to http_request and send_request, including redirect-target validation and two new localhost-detection cases︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.188 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.188
The http_request and send_request tools now refuse requests whose URLs point at the local machine or the dashboard listener, matching the rule already applied to other gated tools.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Each redirect target is now checked against the same local-machine rule before being followed; when a redirect is refused, the tool returns the redirect response with an explanation rather than following it. Not checked yet
The http_request and send_request tools now refuse requests whose URLs point at the local machine or the dashboard listener, matching the rule already applied to other gated tools.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
1 more in Everything
xalgorix adds validation rules for per-scan discord_webhook and api_base fields, and scales Go memory limits automatically with host RAM RAM Working memory that holds programs and data while a computer is running. Its contents usually disappear when power is removed. ︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.187 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.187
└──▷ TRY IT
Send a scan notification to an internal relay on your LAN while confirming the webhook is accepted before the scan starts.
copy
$ curl -s -X POST http://127.0.0.1:9137/api/scan \
-H 'Authorization: Bearer <token>' \
-H 'Content-Type: application/json' \
-d '{"target": "https://staging.example.com", "discord_webhook": "http://192.168.1.50:8080/relay"}'
Cap memory on a 32 GB host to 8 GB instead of the automatic one-third-of-RAM limit, overriding the new scaling formula.
copy
$ XALGORIX_GO_MEM_LIMIT_MB=8192 xalgorix --web
Enforce that per-scan api_base without a matching api_key stays on the endpoint of the stored credential, preventing stored keys from being sent to an arbitrary host named in the request; supplying a per-request api_key replaces the stored key pool for that scan.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Scale the automatic Go soft memory limit with host RAM — hosts up to 12 GB keep 45% of RAM (capped at 4 GB); larger hosts get one-third of RAM with a 32 GB ceiling — overridden in that order by XALGORIX_GO_MEM_LIMIT_MB or GOMEMLIMIT. Not checked yet Resume a refused or still-shutting-down paused scan — a refused resume now answers 409 and leaves the scan resumable, and a scan resuming while still shutting down waits up to 3 seconds before returning 409 with a retry hint. Not checked yet
Enforce that per-scan api_base without a matching api_key stays on the endpoint of the stored credential, preventing stored keys from being sent to an arbitrary host named in the request; supplying a per-request api_key replaces the stored key pool for that scan.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
2 more in Everything
└──▷ BREAKING ON UPGRADE
!
A scan request that sets a `discord_webhook` pointing at the local machine, dashboard listener, link-local, multicast, or cloud metadata address now gets a 400 (or a WebSocket `error` event) instead of proceeding; webhooks already saved in queue state or schedules are checked when sent and a refused one is logged.
Our best guess
!
A scan request that sets `api_base` to a host outside the stored credential endpoint without supplying `api_key` now gets a 400 instead of proceeding.
Our best guess
!
An empty `GOMEMLIMIT=` environment variable is now treated as unset and no longer disables the automatic memory limit.
Our best guess
xalgorix adds moonshot/kimi-latest alias so applications stay on one model string as LiteLLM advances the underlying Kimi model︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
Use moonshot/kimi-latest as a stable model string that automatically tracks the current Kimi model family, available since v1.109.0.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
1 detail in Everything
xalgorix: Use minimax/minimax-latest as a stable model alias that tracks the current MiniMax model, so apps needn't change model strings when LiteLLM upgrades.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
xalgorix: Use stable model strings like baseten/kimi-latest and baseten/glm-latest that automatically track the current model in each family, no code updates needed.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
LiteLLM adds inference-component pinning for AWS SageMaker deployments, with opt-in client override via configurable_clientside_auth_params︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Pin a specific inference component for a deployment so all traffic goes to it, while blocking clients from overriding it.
yaml
copy
model_list:
- model_name: my-llama
litellm_params:
model: sagemaker_chat/<your-endpoint>
extra_headers:
X-Amzn-SageMaker-Inference-Component: <llama-inference-component>
Let clients choose which inference component to target on a shared endpoint by passing model_id in the request body.
yaml
copy
model_list:
- model_name: my-shared-endpoint
litellm_params:
model: sagemaker_chat/<your-endpoint>
model_info:
configurable_clientside_auth_params:
- "model_id"
Send a request targeting a specific inference component when the deployment permits client-side selection via model_id.
copy
$ curl http://0.0.0.0:4000/v1/chat/completions \
-H "Authorization: Bearer $LITELLM_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "my-shared-endpoint", "model_id": "<mistral-inference-component>", "messages": [{"role": "user", "content": "hi"}]}'
Pin a SageMaker inference component at the deployment level via the X-Amzn-SageMaker-Inference-Component header in extra_headers, taking precedence over any model_id the client sends.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Allow clients to select the inference component on a sagemaker_chat/ deployment by listing "model_id" in configurable_clientside_auth_params; requests without model_id fall back to the deployment's own pinned component. Not checked yet LiteLLM now drops any client-supplied X-Amzn-SageMaker-Inference-Component header on sagemaker_chat/ deployments that have a pinned component, preventing client-side override regardless of header casing. Not checked yet
3 details in Everything
└──▷ BREAKING ON UPGRADE
!
Prior to 1.109.0, a `model_id` in the request body would replace a deployment's pinned inference component, letting a client redirect a pinned deployment to a different component. That override no longer works.
Our best guess
xalgorix adds custom video generation provider support via BaseVideoConfig, exposing async POST /v1/videos and GET /v1/videos/{id} endpoints︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ TRY IT
Poll for job completion and then download the finished video file once status is completed .
copy
$ curl 'http://0.0.0.0:4000/v1/videos/video_bGl0ZWxs...' -H 'Authorization: Bearer $LITELLM_API_KEY'
# when status == completed:
curl 'http://0.0.0.0:4000/v1/videos/video_bGl0ZWxs.../content' -H 'Authorization: Bearer $LITELLM_API_KEY' --output video.mp4
Register a custom video backend by subclassing BaseVideoConfig and implementing get_supported_openai_params, map_openai_params, validate_environment, get_complete_url, transform_video_create_request, transform_video_create_response, transform_video_status_retrieve_request, transform_video_status_retrieve_response, transform_video_content_request, and transform_video_content_response.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Submit video generation jobs asynchronously via POST /v1/videos with a prompt and seconds parameter; the response returns immediately with a VideoObject whose status starts as queued. Not checked yet Poll job status via GET /v1/videos/{id} and fetch the finished file via GET /v1/videos/{id}/content, with the encoded id carrying the provider name so no caller-supplied provider is needed. Not checked yet Use encode_video_id_with_provider and extract_original_video_id helpers from base_llm.videos.transformation to embed and recover the provider name in the opaque video ID. Not checked yet Raise NotImplementedError in transform_video_remix_request, transform_video_remix_response, transform_video_delete_request, or transform_video_delete_response when the backend does not support those operations; a provider registered only as a handler returns 400 video generation is not supported for <provider>. Not checked yet
5 details in Everything
xalgorix adds deepseek/<family>-latest model alias support for DeepSeek Pro and Flash families, keeping your model string stable as LiteLLM routes to newer models︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
Pin to a DeepSeek model family rather than a specific version using deepseek/deepseek-pro-latest and deepseek/deepseek-flash-latest, so applications automatically follow model promotions without config changes.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
1 detail in Everything
LiteLLM adds stable bedrock/<family>-latest alias strings (since v1.109.0) so applications keep one model string as LiteLLM moves families to newer models.
New integration Use stable alias strings like bedrock/claude-opus-latest, bedrock/claude-sonnet-latest, bedrock/claude-haiku-latest, bedrock/grok-latest, bedrock/kimi-latest, and bedrock/glm-latest to always route to the current target model in a family without updating your model string when LiteLLM moves to a newer version. Not checked yet
1 detail in Everything
xalgorix adds azure_ai/<family>-latest model family aliases for Azure AI Foundry, keeping one model string as LiteLLM targets newer models︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
Use stable azure_ai/<family>-latest aliases — such as azure_ai/gpt-luna-latest, azure_ai/claude-sonnet-latest, azure_ai/deepseek-pro-latest, and 13 others — so your application string never changes when LiteLLM advances the underlying Azure AI Foundry model.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
1 detail in Everything
xalgorix adds Azure latest model aliases and a realtime speech-to-speech translation endpoint for Azure OpenAI︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ TRY IT
Pin your app to a model family so it automatically follows LiteLLM upgrades without changing your code.
copy
$ curl https://your-proxy/v1/chat/completions \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer <your-key>' \
-d '{"model": "azure/gpt-terra-latest", "messages": [{"role": "user", "content": "Hello"}]}'
Route live speech-to-speech translation through the proxy using the realtime translation endpoint.
copy
$ curl https://your-proxy/v1/realtime/translations \
-H 'Authorization: Bearer <your-key>' \
-d '{"model": "azure/gpt-realtime-translate"}'
Connect to live speech-to-speech translation via the azure/gpt-realtime-translate model on the /v1/realtime/translations proxy endpoint, with WebSocket WebSocket A protocol that upgrades an HTTP connection into a persistent, full-duplex channel, letting a server push data to a client without repeated polling. Cyber tools use it for real-time event streaming and live command output. client, WebRTC WebRTC A browser-native, open standard for real-time peer-to-peer audio, video, and data exchange without plugins. Cyber tools use it to tunnel traffic or exfiltrate data through connections browsers treat as legitimate. , and pricing documented under Realtime translation.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
1 detail in Everything
xalgorix adds thought_signature_in_tool_call_id to shorten Gemini tool call ids and supports_continuation to flag and query mid-answer continuation capability.
Keep Gemini tool call ids short by setting thought_signature_in_tool_call_id: false on a deployment or per request; the thought signature moves to provider_specific_fields.thought_signature instead of being embedded in the id. Not checked yet Check whether a Vertex AI model supports continuation via litellm.supports_continuation() , litellm.get_model_info()['supports_continuation'], or GET /model/info, GET /v1/model/info, and GET /model_group/info. Not checked yet New supported platform Override the cost-map supports_continuation flag for a single deployment by setting supports_continuation: true in the deployment config or Admin UI. Not checked yet
3 details in Everything
xalgorix adds supports_continuation introspection APIs and thought_signature_in_tool_call_id to produce short tool call ids for length-constrained clients.
New supported platform Check continuation support via litellm.supports_continuation() or GET /model/info, GET /v1/model/info, and GET /model_group/info; override per deployment in config.yaml or the Admin UI. Not checked yet Set thought_signature_in_tool_call_id: false in config.yaml or per request to return short ids such as call_436686, with the signature preserved in the tool call object; applies to vertex_ai/ and /responses endpoints with and without streaming. Not checked yet
2 details in Everything
xalgorix: Connect to /v1/realtime/translations via the proxy with openai/gpt-realtime-translate for live speech-to-speech translation with translated audio and text.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ TRY IT
Route live speech-to-speech translation through the LiteLLM proxy without changing your application model string when OpenAI updates the underlying model.
copy
$ curl -X POST http://localhost:4000/v1/realtime/translations \
-H 'Authorization: Bearer <your-litellm-key>' \
-H 'Content-Type: application/json' \
-d '{"model": "openai/gpt-realtime-translate"}'
Pin your application to a model family alias so LiteLLM transparently upgrades to the latest model in that family without code changes.
copy
$ curl -X POST http://localhost:4000/v1/chat/completions \
-H 'Authorization: Bearer <your-litellm-key>' \
-H 'Content-Type: application/json' \
-d '{"model": "openai/gpt-mini-latest", "messages": [{"role": "user", "content": "Hello"}]}'
xalgorix: Control Gemini thought signatures in tool call IDs via thought_signature_in_tool_call_id; set false to keep IDs short (since v1.109.0).︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ TRY IT
Prevent Gemini thought signatures from bloating tool call IDs when using function-calling function-calling A model's ability to answer with a structured request to run a named function or API, which the application executes and returns. It is how agents take actions. workflows on Vertex AI.
copy
$ curl https://your-litellm-proxy/v1/chat/completions \
-H 'Content-Type: application/json' \
-d '{
"model": "gemini/gemini-2.0-flash",
"thought_signature_in_tool_call_id": false,
"messages": [{"role": "user", "content": "What is the weather in London?"}],
"tools": [{"type": "function", "function": {"name": "get_weather", "parameters": {"type": "object", "properties": {"location": {"type": "string"}}}}}]
}'
xalgorix adds audit logging for A2A A2A Agent2Agent, an open protocol started by Google for AI agents from different vendors to discover each other, delegate tasks and exchange results. It complements MCP, which connects an agent to tools. kill switch events, controlled by store_audit_logs or LITELLM_STORE_AUDIT_LOGS︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
Disable A2A kill switch audit logging by setting store_audit_logs to false in config or the LITELLM_STORE_AUDIT_LOGS environment variable to false; audit logging is otherwise on by default when an enterprise license is present.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Kill switch audit log rows record the agent ID, the user and key that fired the switch, the target URL, HTTP method, response status code, truncated response body, and any transport error — but never the kill switch config, auth, custom headers, body, or query params. Not checked yet
2 details in Everything
xalgorix adds file-owns-key config precedence rules since v1.103.0, with 400 errors that name the conflicting key and how to resolve it︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ TRY IT
Attempt to update a config-file-owned key via the API and read the structured error to understand which key is blocked and what to do.
copy
$ curl -X POST https://<proxy-host>/config/field/update \
-H 'Authorization: Bearer <api-key>' \
-H 'Content-Type: application/json' \
-d '{"general_settings": {"store_model_in_db": true}}'
Since v1.103.0, keys declared in the config file are owned by the file — writes to those keys via the Admin UI or POST /config/field/update return a 400 naming the key, its section, any ignored database value in stored_database_values_ignored, and a resolution message telling you to edit the file or remove the key to hand ownership back to the database.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
1 detail in Everything
└──▷ BREAKING ON UPGRADE
!
Since v1.103.0, the config file now takes precedence over the database on any key it declares — the previous behavior (database rows deep-merged over the file, so the database value won) is reversed. If you have keys in both the file and the database, the database values are silently ignored; to restore the old behavior, remove those keys from the file so the database owns them again.
Our best guess
xalgorix adds a Skills Hub API for uploading, importing, discovering, and downloading portable agent skill packages for Codex, Claude Code, and compatible agents︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ TRY IT
Upload a new portable skill package for admin review from your team's local build.
copy
$ zip -r pdf-summarizer.zip pdf-summarizer
curl --fail-with-body https://your-proxy/skill-hub/skills/upload \
-H "Authorization: Bearer $LITELLM_API_KEY" \
-F "file=@pdf-summarizer.zip" \
-F "version=1.0.0"
Before approving a member-submitted skill, trigger a SkillSpector security scan and poll until it completes.
copy
$ curl --fail-with-body -X POST "https://your-proxy/skill-hub/submissions/$SUBMISSION_ID/scan" \
-H "Authorization: Bearer $LITELLM_MASTER_KEY"
# Poll until scan.status is 'completed' or 'failed'
curl --fail-with-body "https://your-proxy/skill-hub/submissions/$SUBMISSION_ID/scan" \
-H "Authorization: Bearer $LITELLM_MASTER_KEY"
Allow unauthenticated agents to discover and download public skills from the registry.
yaml
copy
general_settings:
skill_scanner:
endpoint: https://scanner.example.com/scan
api_key_env: SKILL_SCANNER_API_KEY
use_llm: false
timeout_seconds: 30
litellm_settings:
public_skills_index: true
Upload a portable skill ZIP to POST /skill-hub/skills/upload with a SKILL.md at its root; writable users submit for admin review, proxy admins publish immediately.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Import a public skill from skills.sh/<owner>/<repository>/<skill> or github.com/<owner>/<repository> via POST /skill-hub/skills/import, with optional skill, ref, and directory selection. Not checked yet Search visible skills with GET /skill-hub/skills using query, offset, and limit (1–100); response includes skills, count, and next_offset. Not checked yet Read a skill's original SKILL.md via GET /skill-hub/skills/{name}/content or download its full portable ZIP via GET /skill-hub/skills/{name}/archive. Not checked yet Replace a skill's ZIP with optimistic-concurrency protection by supplying expected_updated_at to PUT /skill-hub/skills/{name}/upload. Not checked yet
+5 more
show less
Browse version history with GET /skill-hub/skills/{name}/versions and fetch a specific version's archive via GET /skill-hub/skills/{name}/versions/{version_id}/archive. Not checked yet Manage submissions as a proxy admin via GET /skill-hub/submissions, PUT /skill-hub/submissions/{submission_id}/approve, and PUT /skill-hub/submissions/{submission_id}/reject with an optional reason body. Not checked yet Trigger an optional pre-approval security scan with POST /skill-hub/submissions/{submission_id}/scan and poll results with the corresponding GET; configure the SkillSpector adapter endpoint and shared secret under general_settings.skill_scanner in the gateway config. Not checked yet Toggle public visibility without deleting a skill using POST /skill-hub/skills/{name}/publish and POST /skill-hub/skills/{name}/unpublish. Not checked yet Enable anonymous public download of skills by setting litellm_settings.public_skills_index: true in the gateway config. Not checked yet
10 details in Everything
xalgorix adds SSO SSO Single sign-on, using one identity provider to sign in to multiple applications. It reduces separate passwords and centralizes account and access management. sign-in for the Admin UI via OIDC OIDC OpenID Connect, a sign-in protocol built on OAuth 2.0. An application uses identity information from a trusted provider instead of maintaining a separate password for each user. , with support for Google, Microsoft, and generic providers︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Wire up Okta or any OIDC provider so the proxy generates the right callback URL and sends users back after sign-in.
.env
copy
PROXY_BASE_URL="https://litellm.yourcompany.com"
GENERIC_CLIENT_ID="<client-id>"
GENERIC_CLIENT_SECRET="<client-secret>"
GENERIC_AUTHORIZATION_ENDPOINT="https://okta.yourcompany.com/oauth2/v1/authorize"
GENERIC_TOKEN_ENDPOINT="https://okta.yourcompany.com/oauth2/v1/token"
GENERIC_USERINFO_ENDPOINT="https://okta.yourcompany.com/oauth2/v1/userinfo"
Inspect what your identity provider identity provider A service that authenticates users and supplies identity information to other applications. It is the sign-in authority behind many single sign-on systems. returns before committing to a production SSO config — useful when user fields are non-standard.
.env
copy
ENABLE_SSO_DEBUG=true
PROXY_BASE_URL="https://litellm.yourcompany.com"
Enable Google SSO by setting GOOGLE_CLIENT_ID and GOOGLE_CLIENT_SECRET environment variables.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Enable Microsoft/Azure SSO by setting MICROSOFT_CLIENT_ID, MICROSOFT_CLIENT_SECRET, and MICROSOFT_TENANT environment variables. Not checked yet Enable any OIDC-compliant provider by setting GENERIC_CLIENT_ID, GENERIC_CLIENT_SECRET, GENERIC_AUTHORIZATION_ENDPOINT, GENERIC_TOKEN_ENDPOINT, and GENERIC_USERINFO_ENDPOINT. Not checked yet Debug SSO responses by setting ENABLE_SSO_DEBUG and registering <PROXY_BASE_URL>/sso/debug/callback; inspect provider fields at <PROXY_BASE_URL>/sso/debug/login on the SSO Debug Information page, then unset ENABLE_SSO_DEBUG when finished. Not checked yet Reach the sign-in flow at <PROXY_BASE_URL>/sso/key/generate and the sign-in callback at <PROXY_BASE_URL>/sso/callback. Not checked yet
+1 more
show less
Read non-standard user fields from your identity provider through userIDPInfo.extra_fields in a custom SSO callback. Not checked yet
6 details in Everything
xalgorix adds workload identity workload identity An identity assigned to software, such as a service or container, rather than a person. It lets the workload authenticate and receive permissions without sharing a user's credentials. (OIDC OIDC OpenID Connect, a sign-in protocol built on OAuth 2.0. An application uses identity information from a trusted provider instead of maintaining a separate password for each user. ) credential references for model providers, eliminating static secrets in cloud deployments.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Run a vLLM model on Google Cloud Run Cloud Run Google's managed container runtime that executes stateless workloads on demand, scaling to zero when idle, so tools can deploy without managing servers or paying for idle capacity. using a workload-identity token instead of a static API key — the proxy fetches and caches the ID token automatically.
yaml
copy
model_list:
- model_name: my-vllm-model
litellm_params:
model: hosted_vllm/my-model
api_base: https://vllm.example.com/v1
api_key: oidc/google/https://vllm.example.com
Route Bedrock calls through a GitHub Actions GitHub Actions GitHub's built-in CI/CD system that runs automated workflows triggered by repository events, letting cyber tools automate testing, building, and releasing without managing separate pipeline infrastructure. OIDC token so CI jobs need no long-lived AWS credentials.
yaml
copy
model_list:
- model_name: bedrock-claude
litellm_params:
model: bedrock/anthropic.claude-3-sonnet-20240229-v1:0
aws_web_identity_token: oidc/github/sts.amazonaws.com
aws_role_name: arn:aws:iam::123456789012:role/my-github-actions-role
aws_session_name: litellm-proxy-session
Restrict which directories the proxy may read OIDC token files from when your tokens live outside the default /var/run/secrets path.
copy
$ export LITELLM_OIDC_ALLOWED_CREDENTIAL_DIRS=/custom/secrets/dir,/another/secrets/dir
Authenticate to Amazon Bedrock without a static secret by setting aws_web_identity_token to an OIDC reference such as oidc/google/<audience>, oidc/github/<audience>, oidc/circleci/, oidc/azure/, oidc/file/<absolute path>, or oidc/env/<VARIABLE_NAME> on the model's credential.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Authenticate to Azure OpenAI without a static secret by setting azure_ad_token to an OIDC reference of the same form, with AZURE_CLIENT_ID and AZURE_TENANT_ID set for the token-exchange application. Not checked yet Override the OIDC token file trust boundary by setting LITELLM_OIDC_ALLOWED_CREDENTIAL_DIRS to a comma-separated list of absolute directories, replacing the defaults of /var/run/secrets and /run/secrets. Not checked yet Override the Google Cloud metadata server host for oidc/google/ references by setting GCE_METADATA_HOST to a host or host:port without a scheme. Not checked yet Set the per-read metadata server timeout for oidc/google/ model credential resolution with GCE_METADATA_TIMEOUT (defaults to 3 seconds when unset or nonpositive). Not checked yet
+1 more
show less
Use a latest model alias such as openai/gpt-luna-latest to track a model family without changing the name in your application. Not checked yet
6 details in Everything
└──▷ BREAKING ON UPGRADE
!
Config file keys for settings now win over database values (the UI and `/config/update` return 400 if you try to change a file-owned key); before v1.103.0 the database value overrode the config file.
Our best guess
xalgorix adds OpenTelemetry OpenTelemetry An open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. baggage propagation that stamps team, key, and request-metadata attributes onto every span in a request︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ TRY IT
Stamp a caller-supplied correlation ID and org/user identity onto every span so traces are filterable by request in your APM APM Application Performance Monitoring — a category of software that collects metrics, traces, and logs from running applications so developers can detect slowdowns and diagnose their root cause. tool.
copy
$ export LITELLM_OTEL_BAGGAGE_METADATA_KEYS=user_api_key_org_id,user_api_key_user_id,user_api_key_alias,user_api_key_end_user_id,requester_ip_address,requester_metadata.correlation_id
curl http://localhost:4000/v1/chat/completions \
-H "Authorization: Bearer $LITELLM_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "gpt-4o", "messages": [{"role": "user", "content": "hi"}], "metadata": {"correlation_id": "corr-123"}}'
Tag all spans for a team with cost-centre and region metadata stored on the team, so spend can be filtered by business unit in your trace backend.
copy
$ export LITELLM_OTEL_BAGGAGE_TEAM_METADATA_KEYS=cost_center,region
curl http://localhost:4000/team/new \
-H "Authorization: Bearer $LITELLM_MASTER_KEY" \
-H "Content-Type: application/json" \
-d '{"team_alias": "ml-platform", "metadata": {"cost_center": "cc-42", "owner": "ml-platform", "region": "eu"}}'
Control which identity attributes are stamped on every span with the LITELLM_OTEL_BAGGAGE_PROMOTED_KEYS env var or callback_settings.otel.baggage_promoted_keys YAML key.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Allowlist specific request-metadata keys to emit as litellm.metadata.<key> span attributes using LITELLM_OTEL_BAGGAGE_METADATA_KEYS or callback_settings.otel.baggage_metadata_keys; nested keys use dotted paths such as requester_metadata.agent.id. Not checked yet Allowlist per-team stored metadata sub-keys to emit as the litellm.team.metadata JSON attribute using LITELLM_OTEL_BAGGAGE_TEAM_METADATA_KEYS or callback_settings.otel.baggage_team_metadata_keys. Not checked yet Datastore spans now automatically carry litellm.metadata.table_name and litellm.metadata.db_operation attributes regardless of allowlist settings. Not checked yet xalgorix now propagates authenticated identity — team, key, model, end-user, and request metadata — as OpenTelemetry baggage so guardrail, router, LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. -call, Redis, and Postgres spans are all filterable by those dimensions without manual re-stamping. Not checked yet
5 details in Everything
└──▷ BREAKING ON UPGRADE
!
Each of `LITELLM_OTEL_BAGGAGE_PROMOTED_KEYS`, `LITELLM_OTEL_BAGGAGE_METADATA_KEYS`, and `LITELLM_OTEL_BAGGAGE_TEAM_METADATA_KEYS` replaces its default list rather than extending it — setting any one of them drops all defaults not re-listed, so every key you want to keep must be listed explicitly.
Our best guess
xalgorix adds session affinity to pin conversation requests to the same deployment, with coverage across LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. API routes and async Router methods︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
Pin every request in a conversation to the deployment that served its first request using session affinity, available as a router pre-call check on routes including /v1/chat/completions, /v1/responses, /v1/embeddings, and async Router methods aresponses and aembedding.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Session affinity now also applies to Bedrock pass-through routes such as /bedrock/model/<model>/converse, but is not carried on relay routes /vllm, /azure, /azure_ai, and /nvidia_nim. Not checked yet Context window Context window The maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. errors now use a matching context_window_fallbacks rule before later fallback orders; when no rule matches the requested group, later orders and generic fallbacks remain available. Not checked yet
Pin every request in a conversation to the deployment that served its first request using session affinity, available as a router pre-call check on routes including /v1/chat/completions, /v1/responses, /v1/embeddings, and async Router methods aresponses and aembedding.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Session affinity now also applies to Bedrock pass-through routes such as /bedrock/model/<model>/converse, but is not carried on relay routes /vllm, /azure, /azure_ai, and /nvidia_nim. Not checked yet
1 more in Everything
xalgorix adds live speech translation support via gpt-realtime-translate on OpenAI and Azure at /v1/realtime/translations︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ TRY IT
Route a real-time speech translation session through xalgorix to gpt-realtime-translate on OpenAI.
copy
$ curl -X POST https://<xalgorix-host>/v1/realtime/translations \
-H 'Authorization: Bearer <api-key>' \
-H 'Content-Type: application/json' \
-d '{"model": "gpt-realtime-translate"}'
xalgorix adds four token-count framing config keys — token_count_framing, token_count_tool_framing, token_count_responses_framing, and token_count_responses_tool_frami︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Prevent 400/404 errors on a Grok 4.1 deployment and get accurate local token counts for tool requests by setting all four framing keys and a prefixed base_model.
yaml
copy
model_list:
- model_name: prod-grok
litellm_params:
model: azure_ai/prod-grok
api_base: https://my-resource.services.ai.azure.com
api_key: os.environ/AZURE_AI_API_KEY
base_model: azure_ai/grok-4-1-fast-reasoning
token_count_framing: 149
token_count_tool_framing: 302
token_count_responses_framing: 180
token_count_responses_tool_framing: 350
Reload built-in framing values for all deployments at runtime after a cost-map update, without restarting the proxy.
copy
$ curl -X POST https://your-litellm-proxy/reload/model_cost_map
Set token_count_tool_framing on a deployment to add a measured token offset to local estimates for requests that include a function tool, covering OpenAI function tools, Anthropic custom tools, and Anthropic built-ins such as web_fetch and code_execution.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Set token_count_responses_framing on a deployment to add a measured offset to local estimates for native Responses API requests with no function tool. Not checked yet Set token_count_responses_tool_framing on a deployment to add a measured offset for native Responses API requests that include a function tool; it takes precedence over token_count_responses_framing and token_count_tool_framing for that request type. Not checked yet Set token_count_framing on a deployment as a catch-all offset applied to every local estimate that a narrower key does not cover; a narrower key replaces it rather than adding to it. Not checked yet Reload the cost map at runtime by calling /reload/model_cost_map to pick up updated framing values for entries such as azure_ai/grok-4-1-fast-reasoning, xai/grok-4.3, deepseek/deepseek-v4-flash, deepseek/deepseek-v4-pro, and fireworks_ai/deepseek-v4p1-flash without restarting the proxy. Not checked yet
+1 more
show less
Force a native Responses API deployment to use chat-completions counting by setting use_chat_completions_api: true on the deployment or in default_litellm_params / router_settings; Azure AI deployments whose base URL is not on .services.ai.azure.com or .openai.azure.com default to this bridge automatically. Not checked yet
Set token_count_tool_framing on a deployment to add a measured token offset to local estimates for requests that include a function tool, covering OpenAI function tools, Anthropic custom tools, and Anthropic built-ins such as web_fetch and code_execution.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Set token_count_responses_framing on a deployment to add a measured offset to local estimates for native Responses API requests with no function tool. Not checked yet Set token_count_responses_tool_framing on a deployment to add a measured offset for native Responses API requests that include a function tool; it takes precedence over token_count_responses_framing and token_count_tool_framing for that request type. Not checked yet Set token_count_framing on a deployment as a catch-all offset applied to every local estimate that a narrower key does not cover; a narrower key replaces it rather than adding to it. Not checked yet Force a native Responses API deployment to use chat-completions counting by setting use_chat_completions_api: true on the deployment or in default_litellm_params / router_settings; Azure AI deployments whose base URL is not on .services.ai.azure.com or .openai.azure.com default to this bridge automatically. Not checked yet
1 more in Everything
xalgorix adds long-decode continuation support via continuation_token and status: requires_continuation for Gemini models that exceed per-request output caps︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Poll a Gemini long-decode interaction to completion when the model's output exceeds Google's per-request cap.
text
copy
# Initial request
response = client.interactions.create(
model="gemini/gemini-4-argon",
messages=[{"role": "user", "content": "Write a complete, rigorous proof of the prime number theorem."}],
generation_config={"max_output_tokens": 65536, "thinking_level": "high"}
)
# Continue until done
steps = list(response.steps)
while response.status == "requires_continuation":
response = client.interactions.create(
model="gemini/gemini-4-argon",
id=response.id,
continuation_token=response.continuation_token
)
steps.extend(response.steps)
# status is now 'completed'
full_text = "".join(c["content"] for c in steps if c.get("content"))
Resume a long decode past Google's per-request output cap by re-sending a request with the interaction id and continuation_token returned when status is 'requires_continuation'; repeat until status is no longer 'requires_continuation'.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Background requests with background: true are continued by Google automatically and require no client-side continuation hops. Not checked yet
2 details in Everything
LiteLLM adds tool-aware model routing via NVIDIA NeMo Switchyard, -latest model aliases, and model deprecation management in its AI Gateway︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
Route coding agent sessions by tool execution history with NVIDIA NeMo Switchyard integration, keeping routine work on an efficient model and escalating to a more capable model when the trajectory turns complex.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Use -latest model aliases in the LiteLLM SDK and AI Gateway to keep the same model name as newer versions arrive. Not checked yet Manage model deprecations in the LiteLLM AI Gateway, including finding affected keys, tracking migration traffic, and notifying owners before provider deadlines. Not checked yet LiteLLM's /v1/messages route now passes Anthropic Claude Code auto mode's server-side safety classifier contract through to the Anthropic API, Bedrock InvokeModel, Bedrock Mantle, and Vertex AI. Not checked yet
Route coding agent sessions by tool execution history with NVIDIA NeMo Switchyard integration, keeping routine work on an efficient model and escalating to a more capable model when the trajectory turns complex.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Use -latest model aliases in the LiteLLM SDK and AI Gateway to keep the same model name as newer versions arrive. Not checked yet Manage model deprecations in the LiteLLM AI Gateway, including finding affected keys, tracking migration traffic, and notifying owners before provider deadlines. Not checked yet
1 more in Everything
LiteLLM adds OpenAI Live sessions on /v1/live/sessions, custom video providers via custom_provider_map, and token_count_framing in model_info for local estimates.
New supported platform Serve custom video providers through /v1/videos by registering them with custom_provider_map. Not checked yet New integration Open OpenAI Live sessions through the new /v1/live/sessions endpoint. Not checked yet Restore a local token-count estimate by setting token_count_framing in model_info. Not checked yet LiteLLM now supports Azure PTU capacity sharing across teams. Not checked yet
4 details in Everything
└──▷ BREAKING ON UPGRADE
!
Remove `lensWorker.enabled` from your Helm values before upgrading — the proxy and its Helm charts no longer package the Lens beta.
Our best guess
!
`default_fallbacks` with a plain string value (e.g. `default_fallbacks: gpt-3.5`) stops the proxy at startup; write it as a list (`default_fallbacks: ["gpt-3.5"]`) instead.
Our best guess
xalgorix now defaults XALGORIX_CONVERGE_FINISH and XALGORIX_CONVERGE_RECONCILE to true, automatically closing scans once coverage obligations are met︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.183 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.183
└──▷ TRY IT
Restore the old indefinite-scan behaviour in a CI environment where you want the scan to run until manually stopped.
copy
$ XALGORIX_CONVERGE_FINISH=false XALGORIX_CONVERGE_RECONCILE=false xalgorix --web
Tune the idle-close window and lead-verification window so long-running scans against large targets wait longer before concluding.
copy
$ XALGORIX_CONVERGE_FINISH=true XALGORIX_CONVERGE_FINISH_IDLE=1000 XALGORIX_CONVERGE_FINISH_LEAD_WINDOW=200 xalgorix --web
Scans now close automatically after XALGORIX_CONVERGE_FINISH_IDLE iterations (default 500) of no new validated findings once the coverage threshold is cleared, with high-confidence open leads getting a XALGORIX_CONVERGE_FINISH_LEAD_WINDOW-iteration (default 100) verification window before closure; set XALGORIX_CONVERGE_FINISH=false to restore the previous behaviour.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Content-discovery artifacts such as VCS VCS Version control system: software that tracks changes to files over time, letting teams collaborate and roll back mistakes. Cyber tools use it to manage source code, configs, and signatures. metadata, config and secret files, logs, and OS cruft now owe only information-exposure checks rather than every injection class. Not checked yet
Scans now close automatically after XALGORIX_CONVERGE_FINISH_IDLE iterations (default 500) of no new validated findings once the coverage threshold is cleared, with high-confidence open leads getting a XALGORIX_CONVERGE_FINISH_LEAD_WINDOW-iteration (default 100) verification window before closure; set XALGORIX_CONVERGE_FINISH=false to restore the previous behaviour.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
1 more in Everything
└──▷ BREAKING ON UPGRADE
!
`XALGORIX_CONVERGE_FINISH` now defaults to `true`; scans that previously ran indefinitely past their coverage threshold will now terminate automatically after idle iterations.
Our best guess
!
`XALGORIX_CONVERGE_RECONCILE` now defaults to `true`; exploratory plan tasks without a vulnerability class that previously kept a fully covered scan open indefinitely will now close once all measurable obligations are done.
Our best guess
Xalgorix now defaults XALGORIX_BOUNDED_CONTEXT to true, archiving tool output and replacing old results with compact stubs to keep agent context bounded︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.182 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.182
└──▷ TRY IT
Run a long autonomous scan without the agent context ballooning — old tool output is archived and swapped for stubs automatically under the new default.
copy
$ XALGORIX_TOOL_ARCHIVE_ACTIVE_WINDOW=10 XALGORIX_TOOL_ARCHIVE_MIN_BYTES=4096 xalgorix --web
Opt out of bounded context for a short scan where you want all raw tool output kept live in the agent conversation at all times.
copy
$ XALGORIX_BOUNDED_CONTEXT=false xalgorix --web
└──▷ BREAKING ON UPGRADE
!
`XALGORIX_BOUNDED_CONTEXT` now defaults to `true`; any setup that depended on the previous default of `false` (unbounded context, full tool output resent every iteration) will behave differently on upgrade — set `XALGORIX_BOUNDED_CONTEXT=false` to restore the old behavior.
Our best guess
xalgorix adds port-pinned scan boundaries and turns on epoch-based coverage convergence by default︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.181 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.181
└──▷ TRY IT
Opt out of epoch-based coverage convergence when running a scan in an environment where the previous unbounded behaviour is required.
copy
$ XALGORIX_EPOCH_COVERAGE=false xalgorix --web
Scans are now limited to the port specified in the target URL — requests, port scans, and findings on any other port of that host are refused, so a scan of https://example.com:9000 stays on :9000 only; service enumeration must name the pinned port, for example -p 9000.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Single-target scans no longer enumerate subdomains — subdomain discovery, takeover checks, and sibling subdomain content discovery are skipped; a www. variant still counts as the target. Not checked yet The XALGORIX_EPOCH_COVERAGE environment variable now defaults to true, measuring coverage against a frozen surface snapshot that advances in bounded epochs so scans can close on large or still-growing attack surfaces attack surfaces The exposed systems, interfaces and access points an attacker could try to use. It includes internet-facing services as well as identities, software and other reachable resources. without dropping any discovered endpoint. Not checked yet
Scans are now limited to the port specified in the target URL — requests, port scans, and findings on any other port of that host are refused, so a scan of https://example.com:9000 stays on :9000 only; service enumeration must name the pinned port, for example -p 9000.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
2 more in Everything
└──▷ BREAKING ON UPGRADE
!
`XALGORIX_EPOCH_COVERAGE` now defaults to `true`; any setup that relied on the previous off-by-default behaviour must set `XALGORIX_EPOCH_COVERAGE=false` explicitly to restore it.
Our best guess
!
A port in the target URL now restricts the scan to that port only — scans that previously relied on automatic enumeration and coverage of additional ports on the same host will no longer produce those findings.
Our best guess
!
Single-target scans of a registrable domain no longer enumerate or cover subdomains — workflows that depended on automatic subdomain discovery and takeover checks in single-target mode will no longer trigger them.
Our best guess
Xalgorix now routes requests with the full model ID intact when XALGORIX_LLM_PROVIDER names a gateway or inference host︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.180 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.180
Gateway and inference-host providers named in XALGORIX_LLM_PROVIDER (OpenRouter OpenRouter A unified API gateway that routes requests to dozens of LLM providers under one endpoint, letting cyber tools swap or compare models without rewriting provider-specific integration code. , Groq, Together, vLLM, LM Studio, and others) now receive requests with the model ID unchanged — a slash-prefixed ID such as google/gemini-3-pro or qwen/qwen3-32b is no longer stripped before forwarding.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Built-in multi-vendor providers such as OpenRouter and Together now use their catalog base URL automatically when XALGORIX_API_BASE is unset, instead of falling back to a generic default host. Not checked yet Xalgorix now logs a one-time message when a provider setting overrides a model-ID prefix, so a stale XALGORIX_LLM_PROVIDER or XALGORIX_API_BASE setting is easy to spot. Not checked yet
Gateway and inference-host providers named in XALGORIX_LLM_PROVIDER (OpenRouter OpenRouter A unified API gateway that routes requests to dozens of LLM providers under one endpoint, letting cyber tools swap or compare models without rewriting provider-specific integration code. , Groq, Together, vLLM, LM Studio, and others) now receive requests with the model ID unchanged — a slash-prefixed ID such as google/gemini-3-pro or qwen/qwen3-32b is no longer stripped before forwarding.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
2 more in Everything
└──▷ BREAKING ON UPGRADE
!
Built-in multi-vendor providers (OpenRouter, Together, and others with no legacy default) now use their catalog base URL when `XALGORIX_API_BASE` is unset — setups that relied on the previous generic fallback host will now send traffic to a different base URL.
Our best guess
xalgorix adds NVIDIA NeMo Switchyard as a tool-trajectory-aware router strategy, configurable via switchyard_router_config YAML, with per-session state, auditable spend logs, and custom tool-class vocabularies.
Configure Switchyard routing via switchyard_router_config in model_list YAML, specifying capable and efficient model names, with an optional LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. or decision-model judge for inconclusive turns. Not checked yet Connect Claude Code to a Switchyard router by setting ANTHROPIC_BASE_URL, ANTHROPIC_AUTH_TOKEN, and ANTHROPIC_MODEL to the router name, with no client-side code changes required. Not checked yet New integration xalgorix now supports NVIDIA NeMo Switchyard as an auto_router/switchyard strategy that routes each agent turn based on the session's rolling tool-call history — escalating to a capable model on failures and stepping back down on recovery — rather than on prompt content alone. Not checked yet Custom tool vocabularies let internal or MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. tools be declared as observe, mutate, plan, or new to contribute to Switchyard's trajectory signals alongside built-in classifications. Not checked yet Switchyard routing decisions are recorded in spend logs with the selected model and evidence, enabling cost and quality analysis per session. Not checked yet
5 details in Everything
xalgorix adds -latest model aliases so apps automatically track the newest model version without code changes︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Call the latest OpenAI model in a family without pinning a version — xalgorix will route to the current best model behind the alias.
python
copy
from litellm import completion
response = completion(
model="openai/gpt-luna-latest",
messages=[{"role": "user", "content": "Say hello."}],
)
Use auto-updating -latest model aliases — such as openai/gpt-luna-latest, anthropic/claude-sonnet-latest, or deepseek/deepseek-pro-latest — in the Python SDK and AI Gateway so the model behind the name updates without touching application code.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Supported -latest aliases span OpenAI (gpt-luna-latest, gpt-sol-latest, gpt-terra-latest, gpt-astra-latest, gpt-mini-latest), Anthropic (claude-opus-latest, claude-sonnet-latest, claude-haiku-latest, claude-fable-latest), xAI (grok-latest), Moonshot (kimi-latest), Z.AI (glm-latest, glm-flash-latest), DeepSeek (deepseek-pro-latest, deepseek-flash-latest), DashScope (qwen-max-latest, qwen-flash-latest), and MiniMax (minimax-latest). Not checked yet
2 details in Everything
xalgorix adds model deprecation alerts, scheduled email reminders, and per-key usage tracking to help prevent outages during model migrations︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
Enable model deprecation email reminders by turning on Model Deprecations under Logging & Alerts → Email Alerts → Email Notifications in the UI; xalgorix schedules two reminders at 30 days and 1 day before the provider's deprecation date.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Review per-key traffic during a migration using the Review impact view, which shows a daily requests chart broken down by key alongside affected keys, teams, owners, and tags. Not checked yet View model deprecation alerts — including provider, model, deprecation date, requests in the last 30 days, and last-seen traffic — under Observability → Alerts → Model deprecations in the UI. Not checked yet Mark a migration complete with Mark resolved so the alert moves to Resolved; new requests to the old model automatically reopen it. Not checked yet
4 details in Everything
xalgorix adds authentication-bypass detection, JSON body field discovery, and body injection for SQL and template injection checks︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.178 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.178
Detect authentication-bypass vulnerabilities with the new verify_authbypass check, which re-requests protected routes unauthenticated using equivalent spellings (case fold, percent-encoding, trailing slash, dot and matrix separators) and reports confirmed bypasses under the new auth-bypass report class (CWE CWE Common Weakness Enumeration, a catalog of recurring software and hardware weakness types. It describes classes of mistakes, such as injection, rather than individual vulnerability instances. -287/288/425) at medium or high severity.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Discover which JSON body fields a POST/PUT/PATCH endpoint expects with the new probe_body_contract tool, which classifies each field as required, recognized, rejected as unknown, or type-validated by reading the app's own validation errors. Not checked yet Inject into JSON or form request bodies in addition to query strings with verify_sqli and verify_ssti, using the new options inject_in, body_type, and base_body. Not checked yet Detect error-based SQL injection SQL injection An attack that makes an application treat untrusted input as part of a database command. It can expose or modify data by changing what the query does. on PostgreSQL and SQL Server through expanded engine-specific database error signatures in verify_sqli. Not checked yet
Detect authentication-bypass vulnerabilities with the new verify_authbypass check, which re-requests protected routes unauthenticated using equivalent spellings (case fold, percent-encoding, trailing slash, dot and matrix separators) and reports confirmed bypasses under the new auth-bypass report class (CWE CWE Common Weakness Enumeration, a catalog of recurring software and hardware weakness types. It describes classes of mistakes, such as injection, rather than individual vulnerability instances. -287/288/425) at medium or high severity.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Discover which JSON body fields a POST/PUT/PATCH endpoint expects with the new probe_body_contract tool, which classifies each field as required, recognized, rejected as unknown, or type-validated by reading the app's own validation errors. Not checked yet
2 more in Everything
xalgorix adds proven confirmations for SQL injection SQL injection An attack that makes an application treat untrusted input as part of a database command. It can expose or modify data by changing what the query does. , template injection and XXE XXE XML External Entity injection, a class of attack against XML parsers that tricks them into fetching external resources or local files, letting an attacker read sensitive data or trigger server-side request forgery. , case-sensitive dedup keys, and a corrected scan-completion lifecycle︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.177 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.177
xalgorix's SQL injection, template injection and in-band XXE verifiers now mark confirmations as proven — including when a confirmation merges into an existing ledger entry — so a scan can no longer end with a confirmed injection unreported.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Ledger evidence now records which engine verifier wrote it, so model-supplied text such as not confirmed is no longer treated as verifier proof. Not checked yet Per-type severity caps now apply after CVSS CVSS Common Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. scoring, so a model-supplied vector can no longer raise capped finding types such as clickjacking or missing headers to medium or high. Not checked yet Out-of-band and cross-identity signals now stay open for the model to conclude and are never auto-closed as exhausted, with the verification window waiting on them. Not checked yet Ledger and finding-dedup keys now preserve path case — lowercasing only scheme and host — so a finding on /Api/... is no longer merged with one on /api/..., keeping case-variant authentication bypasses as separate findings. Not checked yet
+2 more
show less
Scans that exhaust or saturate their plan now let the no-new-finding tail decide termination, so the completion assessment, the record of untested work, and the verification window are no longer skipped by the stuck-loop limit. Not checked yet Rate-limit, quota and provider-error retries no longer consume idle limits or the verification window, and a scan paused by a provider outage keeps its idle budget. Not checked yet
xalgorix's SQL injection, template injection and in-band XXE verifiers now mark confirmations as proven — including when a confirmation merges into an existing ledger entry — so a scan can no longer end with a confirmed injection unreported.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
6 more in Everything
Xalgorix adds Runcrate as a built-in LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. provider and now reports total LLM cost across all agents at the end of each scan.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.176 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.176
Xalgorix now adds up LLM costs reported by the provider (such as OpenRouter OpenRouter A unified API gateway that routes requests to dozens of LLM providers under one endpoint, letting cyber tools swap or compare models without rewriting provider-specific integration code. 's usage.cost) across the scan and every subagent, and shows the total in the CLI's end-of-scan usage note.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Xalgorix adds Runcrate — a chat-completions API for open-weight models — to the built-in provider catalog, with model discovery via /v1/models. Not checked yet
2 details in Everything
Xalgorix adds a bounded verification window and accurate arjun rate-limit options to prevent premature scan closure and failed parameter discovery︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.175 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.175
└──▷ TRY IT
Tune scan endurance on hardened targets where findings are rare — raise the idle tail and give the agent more time to verify promising leads before the scan closes.
copy
$ XALGORIX_CONVERGE_FINISH=true XALGORIX_CONVERGE_FINISH_IDLE=1000 XALGORIX_CONVERGE_FINISH_LEAD_WINDOW=200 xalgorix --web
Persist tuned convergence settings across restarts by writing them to ~/.xalgorix.env so every scan inherits them without re-exporting each time.
~/.xalgorix.env
copy
XALGORIX_CONVERGE_FINISH=true
XALGORIX_CONVERGE_FINISH_IDLE=800
XALGORIX_CONVERGE_FINISH_LEAD_WINDOW=150
Set XALGORIX_CONVERGE_FINISH_IDLE to control the no-finding tail length before a scan finalizes (default raised from 150 to 500 iterations) when XALGORIX_CONVERGE_FINISH is on.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Use XALGORIX_CONVERGE_FINISH_LEAD_WINDOW to give the agent a bounded verification window (default 100 iterations) to verify high-confidence unverified leads before the scan closes. Not checked yet Scans closed while work remained unexecuted now record completion: partial with stop_reason: untested_at_finish, listing unrun tasks as untested_at_finish rather than not applicable , and the finish message reports how much of the plan and route surface was actually tested. Not checked yet The request-rate rewriter now gives arjun its real -t and --rate-limit options, maps commonly guessed flags (--threads, --chunk, --delay, --stable-sleep, --timeout, -oJ) to their real equivalents, and limits rewriting to commands that actually invoke the tool. Not checked yet
Set XALGORIX_CONVERGE_FINISH_IDLE to control the no-finding tail length before a scan finalizes (default raised from 150 to 500 iterations) when XALGORIX_CONVERGE_FINISH is on.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Use XALGORIX_CONVERGE_FINISH_LEAD_WINDOW to give the agent a bounded verification window (default 100 iterations) to verify high-confidence unverified leads before the scan closes. Not checked yet
2 more in Everything
xalgorix v4.6.173 adds parser support for DSML-namespaced tool-call markup so model invocations are no longer silently dropped.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.173 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.173
Supports DSML-namespaced tool-call markup (<|DSML|invoke name="…"> / <|DSML|parameter …>) so scans no longer stall at the no tool call stop limit when a model emits DSML-tagged invocations.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
1 detail in Everything
LiteLLM adds abliterated model providers Refuseless and Abliteration, routable via existing gateway config with access controls intact.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Add both Refuseless and Abliteration providers to your gateway for red-teaming or security research workflows that require uncensored model responses.
config.yaml
copy
model_list:
- model_name: refuseless-chat
litellm_params:
model: refuseless/glm-5.3-abliterated
api_key: os.environ/REFUSELESS_API_KEY
- model_name: abliteration-chat
litellm_params:
model: abliteration/abliterated-model-large-v2
api_key: os.environ/ABLITERATION_API_KEY
Send a chat completion to the Refuseless abliterated GLM model via the LiteLLM gateway, using an existing virtual key.
copy
$ curl "$LITELLM_BASE_URL/v1/chat/completions" \
-H "Authorization: Bearer $LITELLM_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "refuseless-chat",
"messages": [{"role": "user", "content": "Explain how DNS works."}],
"reasoning_effort": "low"
}'
Adds refuseless/ prefix support and REFUSELESS_API_KEY env var to route requests to Refuseless abliterated models (glm-5.3-abliterated, glm-5.3-flash-abliterated, deepseek-v4.1-flash-abliterated) and original-weight models (deepseek-v4.1-flash-fast, minimax-m3.1-flash) through the LiteLLM gateway.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds abliteration/ prefix support and ABLITERATION_API_KEY env var to route requests to Abliteration models (abliterated-model, abliterated-model-large-v2, abliterated-model-large-flash) — with image input supported on Base and Large Flash variants. Not checked yet Adds both providers to model_list in config.yaml under litellm_params, keeping existing gateway authentication, budgets, rate limits, and routing policies in place. Not checked yet Supports streaming, tool calls, structured output structured output Forcing a model's response to match a given format or JSON schema, so software can parse it reliably instead of hoping the text is well formed. , and cached-token pricing for abliterated model requests routed through the gateway. Not checked yet
4 details in Everything
LiteLLM gains Day 0 GPT-6.1 Terra support with automatic Responses API routing for tool calls, a new POST /reload/model_cost_map endpoint, and reasoning_effort parameter support.
New integration Adds openai/gpt-6.1-terra as a supported model in config.yaml under model_list, available on v1.76.0 and above. Not checked yet Automatically routes /chat/completions requests with tools to /v1/responses when reasoning is enabled for GPT-6.1 Terra. Not checked yet Adds POST /reload/model_cost_map endpoint (or UI Reload Price Data under Models + Endpoints) to pull updated pricing without a proxy restart. Not checked yet Supports the reasoning_effort parameter (values: none, low, medium, max; default medium) for GPT-6.1 Terra via both the proxy and the Python SDK. Not checked yet
4 details in Everything
Adds opt-in native tool calling tool calling A model's ability to answer with a structured request to run a named function or API, which the application executes and returns. It is how agents take actions. via XALGORIX_NATIVE_TOOL_CALLS for llama.cpp, Ollama, and vLLM backends, replacing fragile XML tag parsing.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.163 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.163
└──▷ TRY IT
Enable native tool calling when using an Ollama backend to allow models not trained on XML to drive tools reliably during autonomous scans.
copy
$ XALGORIX_NATIVE_TOOL_CALLS=true xalgorix --web
xalgorix v4.6.160 adds target-wide finding deduplication classes and two opt-in env-var flags for stuck-scan recovery and leaner tool schemas.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.160 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.160
└──▷ TRY IT
Keep a long autonomous scan alive when one task stalls — useful in CI where an unresponsive endpoint would otherwise abort the entire run.
copy
$ XALGORIX_ADVANCE_PAST_STUCK=true XALGORIX_MAX_STUCK_ADVANCES=5 xalgorix --web
Reduce per-request prompt size when using a model with a tighter context window context window The maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. , without losing any tool or parameter coverage.
copy
$ XALGORIX_LEAN_TOOL_SCHEMA=true xalgorix --web
Adds XALGORIX_ADVANCE_PAST_STUCK environment variable (default off) to skip un-completable tasks and continue the rest of the scan plan, bounded by XALGORIX_MAX_STUCK_ADVANCES (default 3) consecutive no-progress advances.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds XALGORIX_LEAN_TOOL_SCHEMA environment variable (default off) to strip illustrative example spans from tool/parameter descriptions, shrinking the per-request prompt while preserving every tool, parameter and core instruction. Not checked yet Adds debug_disclosure finding class (stack-trace / debug-page / verbose-exception keywords; CWE CWE Common Weakness Enumeration, a catalog of recurring software and hardware weakness types. It describes classes of mistakes, such as injection, rather than individual vulnerability instances. -209) so error-verbosity findings are classified and collapsed to one finding per target instead of being re-reported on every path. Not checked yet Adds user_enumeration finding class (CWE-203/204) so account-enumeration reports on the same endpoint deduplicate. Not checked yet Collapses server-wide issue classes — debug/stack-trace disclosure, missing security headers, CORS CORS A browser mechanism that lets a server declare which foreign origins may read its HTTP responses, enforced by the browser to block unauthorized cross-site requests from JavaScript. policy, version/technology disclosure, and clickjacking — to one finding per target instead of re-reporting on every path. Not checked yet
Adds XALGORIX_ADVANCE_PAST_STUCK environment variable (default off) to skip un-completable tasks and continue the rest of the scan plan, bounded by XALGORIX_MAX_STUCK_ADVANCES (default 3) consecutive no-progress advances.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds XALGORIX_LEAN_TOOL_SCHEMA environment variable (default off) to strip illustrative example spans from tool/parameter descriptions, shrinking the per-request prompt while preserving every tool, parameter and core instruction. Not checked yet Adds debug_disclosure finding class (stack-trace / debug-page / verbose-exception keywords; CWE CWE Common Weakness Enumeration, a catalog of recurring software and hardware weakness types. It describes classes of mistakes, such as injection, rather than individual vulnerability instances. -209) so error-verbosity findings are classified and collapsed to one finding per target instead of being re-reported on every path. Not checked yet Adds user_enumeration finding class (CWE-203/204) so account-enumeration reports on the same endpoint deduplicate. Not checked yet
1 more in Everything
Xalgorix v4.6.159 adds XALGORIX_PLAN_COMPLETE_IDLE to cap idle scan iterations and recovers mis-tagged LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. tool calls to cut wasted token usage.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.159 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.159
└──▷ TRY IT
Lower the idle-iteration cap for fast CI scans where you want earlier finalization once the plan is done and no new findings appear.
copy
$ XALGORIX_PLAN_COMPLETE_IDLE=100 xalgorix --web
Persist a tighter idle cap for self-hosted installs that store config in ~/.xalgorix.env.
~/.xalgorix.env
copy
XALGORIX_PLAN_COMPLETE_IDLE=200
Adds XALGORIX_PLAN_COMPLETE_IDLE environment variable (default 600) to bound the no-new-findings tail after a structured plan is exhausted, finalizing scans instead of re-testing already-covered surface indefinitely; any new finding resets the counter.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Recovers LLM tool calls where the model emits <parameter=NAME> instead of <function=NAME>, avoiding the wasted turn and full-context retry that previously occurred when the call was discarded as malformed. Not checked yet
Adds XALGORIX_PLAN_COMPLETE_IDLE environment variable (default 600) to bound the no-new-findings tail after a structured plan is exhausted, finalizing scans instead of re-testing already-covered surface indefinitely; any new finding resets the counter.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
1 more in Everything
LiteLLM proxy gains Kubernetes pod discovery for direct per-pod routing with session-affinity and a new Admin UI pods card.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Route inference traffic directly to individual vLLM pods behind a Kubernetes headless service, enabling per-session prompt-cache reuse.
litellm_config.yaml
copy
model_list:
- model_name: llama
litellm_params:
model: openai/fake
api_base: http://headless.gpu.svc.cluster.local:8000/v1
api_key: os.environ/VLLM_API_KEY
kubernetes_pod_discovery: true
Inspect which pod IPs a model currently resolves to — useful for verifying pod discovery is working before sending traffic.
copy
$ curl -X GET 'https://<proxy-host>/model/<model_id>/kubernetes_pods' \
-H 'Authorization: Bearer <admin-api-key>'
Adds kubernetes_pod_discovery: true config key to enable direct per-pod round-robin routing when the model base URL is a Kubernetes headless service, refreshing pod IPs on the interval set by KUBERNETES_POD_DISCOVERY_REFRESH_INTERVAL_SECONDS.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds session-affinity to Kubernetes pod discovery: requests carrying litellm_session_id or session_id in metadata are pinned to one pod so the inference server can reuse its prompt cache; pod IPs unused for KUBERNETES_POD_DISCOVERY_IDLE_EVICTION_SECONDS are evicted from the cache. Not checked yet Adds GET /model/{model_id}/kubernetes_pods endpoint (admin-only) to retrieve the pod IPs a model currently resolves to. Not checked yet Adds a Kubernetes pods card to the Admin UI model page (Models + Endpoints > select model) showing the service host, port, and each ready pod IP — visible to proxy admins only. Not checked yet Exposes the kubernetes_pod_discovery setting in Admin UI at Add Model > Advanced Settings > Kubernetes pod discovery . Not checked yet
+1 more
show less
Adds model alias routing so all requests for gpt-6.1-sol are forwarded to models configured as gpt-6-luna. Not checked yet
6 details in Everything
└──▷ BREAKING ON UPGRADE
! Only http:// base URLs are rewritten to pod IPs; https:// bases are left unchanged because connecting to a bare IP breaks TLS hostname checks.! If HTTP_PROXY is set and NO_PROXY exempts the service host but not each pod IP, LiteLLM keeps the service hostname instead of pod IPs and logs a warning, because HTTP clients do not match pod IPs against CIDR CIDR A notation for a block of IP addresses, such as 10.0.0.0/24. The number after the slash says how many leading address bits identify the network. entries in NO_PROXY.
LiteLLM proxy gains self-serve budget policy and route-allowlist defaults for virtual keys︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Allow key owners to add tighter budget windows or lower caps themselves, without filing an admin request every time.
config.yaml
copy
general_settings:
self_serve_budget_policy: lower_only
With lower_only active, a key owner tightens their own key by adding a 24h sub-window under an existing 7d budget.
copy
$ curl http://localhost:4000/key/update \
-H "Authorization: Bearer $OWNER_KEY" \
-d '{"key": "'$OWNER_KEY'", "budget_limits": [{"budget_duration": "7d", "max_budget": 50}, {"budget_duration": "24h", "max_budget": 5}]}'
Adds general_settings.self_serve_budget_policy in config.yaml (or via POST /config/update) with three values — disabled (default), lower_only, and ceiling — letting key owners adjust their own budget_limits on POST /key/update without requiring an admin.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds a route-allowlist default for new virtual keys, configured via llm_api_routes (e.g. ["llm_api_routes", "/key/spend/report"]), so proxy admins can grant specific route groups to newly created keys regardless of the owner's role. Not checked yet
2 details in Everything
xalgorix v4.6.154 adds local LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. endpoint support and prevents custom env var overwrites
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.154 https://github.com/xalgord/xalgorix.git
$ git checkout v4.6.154
Settings → LLM now exposes an endpoint field for local runtimes (Ollama / LM Studio) Not checked yet Re-saving the same provider no longer overwrites a custom XALGORIX_API_BASE Not checked yet
2 details in Everything
xalgorix v4.6.150 improves Go install guidance and assessment coverage logic.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.150 https://github.com/xalgord/xalgorix.git
$ git checkout v4.6.150
└──▷ USE IT
Install the latest version of xalgorix directly from the Go module proxy.
shell
copy
go install github.com/xalgord/xalgorix/v4/cmd/xalgorix@latest
Improves assessment coverage by preventing external references or local artifacts from blocking it, counting rejected plan updates once per agent turn. Not checked yet
1 detail in Everything
v4.6.147 adds plain-text endpoint list uploads to scanner context and improves token-usage tracking across scan sessions.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.147 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.147
Accepts plain-text endpoint lists in uploaded scanner context, enabling practitioners to feed raw target lists directly into a scan session.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Records and reconciles token usage at session boundaries for more accurate scan totals. Not checked yet Reduces repeated agent context while preserving tool schemas and required guidance, lowering noise in long autonomous scans. Not checked yet
Accepts plain-text endpoint lists in uploaded scanner context, enabling practitioners to feed raw target lists directly into a scan session.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
2 more in Everything
Route discovery now filters probe artifacts and merges observations for relative and parameterized routes so coverage reflects real application inputs.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.145 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.145
xalgorix v4.6.134 enhances coordinator instance handling for child scans
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.134 https://github.com/xalgord/xalgorix.git
$ git checkout v4.6.134
Wildcard child scan details retain their own saved findings, events, usage, plan, and outcome when sharing a coordinator instance. Not checked yet
1 detail in Everything
xalgorix v4.6.132 adds scan recovery checkpoints and XALGORIX_MAX_NON_PROGRESS config to resume interrupted assessments
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.132 https://github.com/xalgord/xalgorix.git
$ git checkout v4.6.132
Adds XALGORIX_MAX_NON_PROGRESS environment variable to control the non-progress limit for scan recovery loops Not checked yet Preserves completed work, scan age, finding references, and verified upgrades across restart via executable assessment checkpoints Not checked yet
2 details in Everything
xalgorix v4.6.128 adds one-click environment configuration backup and restore via Settings → Environment
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.128 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.128
└──▷ HOW TO FIND IT
Backup the current environment configuration for migration.
📍 In the console, go to Settings › Environment and use the Backup & migration card to export settings as JSON.
Restore a previous configuration bundle into a fresh install.
📍 In the console, go to Settings › Environment and use the Backup & migration card to import the JSON file.
Adds a Backup & migration card in Settings → Environment for one-click export and import of installation settings. Not checked yet Export downloads the complete configuration as a JSON file including all set environment variables and LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. auth profiles. Not checked yet Import merges a configuration bundle into a fresh install with atomic application, skipping unknown keys. Not checked yet
3 details in Everything
Adds /api/pool endpoints for scoped machine access to LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. API key count and masked values.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.127 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.127
└──▷ TRY IT
Check current LLM API key count and masked IDs without exposing secrets in automation scripts.
copy
$ curl -H 'Authorization: Bearer <machine_token>' https://xalgorix.example.com/api/pool
Rotate the entire LLM API key pool for the xalgorix cyber tool via an authorized machine client.
copy
$ curl -X POST -H 'Authorization: Bearer <machine_token>' -H 'Content-Type: application/json' -d '{"keys": ["sk-new1", "sk-new2"]}' https://xalgorix.example.com/api/pool
Adds GET /api/pool endpoint that returns only the key count and per-key masked values (never secrets) for the LLM API key pool Not checked yet Adds POST /api/pool endpoint that replaces the entire key pool, where the first key becomes the primary Not checked yet
2 details in Everything
xalgorix v4.6.120 introduces single-agent mode where root owns the full methodology with no specialists.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.120 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.120
v4.6.119 introduces applicability-driven orchestration and rebuilds cloud/API security skills as offensive playbooks
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.119 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.119
Adds applicability-driven orchestration with new directives, typed dispositions, and honest completion logic Not checked yet Rebuilds cloud-security skills into an offensive cloud pentest knowledge layer Not checked yet Consolidates API security skills into evidence-driven offensive playbooks Not checked yet Rebuilds application-security skills as a black-box application reasoning layer Not checked yet
4 details in Everything
Adds XALGORIX_DISABLED_SPECIALISTS environment variable to control specialist toggles in the dashboard.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.116 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.116
xalgorix v4.6.112 adds a stepwise phase chain with always-present discovery for agent execution
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.112 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.112
Adds deterministic recon-first sequencing to the agent for consistent coverage
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.102 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.102
xalgorix v4.6.99 adds XALGORIX_DISABLE_AUTO_DELEGATE env var to settings UI
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.99 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.99
└──▷ HOW TO FIND IT
Disable auto-delegation via the settings interface by locating the XALGORIX_DISABLE_AUTO_DELEGATE control.
📍 In the settings UI, find and set XALGORIX_DISABLE_AUTO_DELEGATE to true
Adds feat: LLM API key pool with rate-limit rotation feature for managing LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. keys.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.93 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.93
Embeds launch overview video and compresses sponsor banner image in documentation Not checked yet
1 detail in Everything
xalgorix v4.6.92 adds Realbench multi-CVE benchmark suite with deterministic verification primitives and a Kubernetes Helm chart for deployment.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.92 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.92
└──▷ TRY IT
Deploy xalgorix on Kubernetes using the new Helm chart with hardened defaults and optional ingress configuration.
copy
$ helm install xalgorix ./deploy/chart --set serviceAccount.automount=false --set env.raw='{"DB_HOST":"postgres"}'
Configure the Helm chart to use an existing Kubernetes secret for authentication instead of plain text values.
copy
$ helm install xalgorix ./deploy/chart --set auth.existingSecret=my-xalgorix-secret
verify_path_traversal, verify_timing, discover_client_routes, and verify_path_template_xss as new deterministic verification primitives in the realbench module Not checked yet Realbench OAST OAST Out-of-band Application Security Testing: a technique for detecting vulnerabilities by having a target system make DNS or HTTP callbacks to an attacker-controlled server, revealing flaws that produce no visible in-band response. execution attribution logic distinguishing os-command/runtime-api from fetch-only callbacks for RCE RCE Remote code execution, the ability to run attacker-chosen code on another system through a vulnerability. Its impact depends on the permissions and access that code receives. proof Not checked yet Helm chart at deploy/chart enabling Kubernetes deployment with optional HPA/Ingress/Gateway API support Not checked yet
3 details in Everything
xalgorix v4.6.91 adds dashboard-configurable token-saver flags XALGORIX_BOUNDED_CONTEXT and XALGORIX_ROLE_SCOPED_TOOLS with hot-apply.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.91 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.91
└──▷ HOW TO FIND IT
Enable bounded context to archive aged tool results and fetch them via read_tool_output when the context window context window The maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. reaches the threshold.
📍 Navigate to Settings → Environment in the dashboard, set Bounded working context (token saver) to true, adjust Bounded context: active window (default 8) and Bounded context: archive threshold (bytes) (default 1500).
Enable role-scoped tools to exclude foreign tool documentation from specialist prompts, reducing prompt size.
📍 Navigate to Settings → Environment in the dashboard and toggle Role-scoped specialist tools (token saver) (XALGORIX_ROLE_SCOPED_TOOLS) to true.
Monitor the impact of token savers on scan efficiency using the Token usage tab metrics.
📍 View average_prompt_tokens and resend_amplification_ratio in the scan's Token usage tab after enabling flags.
Adds Bounded working context (token saver) (XALGORIX_BOUNDED_CONTEXT) setting to reduce input tokens by 40–60% via aged message archival. Not checked yet Adds Role-scoped specialist tools (token saver) (XALGORIX_ROLE_SCOPED_TOOLS) setting to reduce input tokens by 10–20% by withholding role-foreign tool docs. Not checked yet Enables hot-application of Environment settings in the dashboard without restarting the runtime. Not checked yet
3 details in Everything
Adds XALGORIX_BOUNDED_CONTEXT and XALGORIX_ROLE_SCOPED_TOOLS env vars for token optimization in xalgorix v4.6.90
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.90 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.90
Adds the read_tool_output tool to retrieve full raw outputs from the bounded working context archive. Not checked yet Adds XALGORIX_BOUNDED_CONTEXT=1 environment variable to enable aged tool-result message compaction and persistent output storage in <ScanDir>/tool-outputs/. Not checked yet Adds XALGORIX_ROLE_SCOPED_TOOLS=1 environment variable to withhold role-foreign tool documentation from delegated specialists' prompts. Not checked yet Ensures deterministic serialization of the tool schema instead of randomizing it per build. Not checked yet
4 details in Everything
xalgorix v4.6.89 adds GET /api/scans/{id}/token-usage endpoint and a Token usage dashboard tab for scan diagnostics.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.89 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.89
GET /api/scans/{id}/token-usage endpoint returns per-request token attribution, context growth analysis (first/last/median/p95/max prompt tokens, prompt growth ratio, cumulative prompt tokens, resend-amplification heuristic), and breakdowns by agent and active-context source. Not checked yet Persistent token diagnostics saved as token-usage.jsonl and token-usage.json under the scan directory to survive server restarts and resumed scans. Not checked yet [token-analysis] summary logged at scan completion for quick post-run review. Not checked yet
3 details in Everything
xalgorix v4.6.85 adds zai-glm-latest support and persists scan tokens across restarts
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.85 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.85
Adds zai-glm-latest (GLM 5.3) model support via Mistral provider for use with OpenAI-compatible API responses. Not checked yet Persisting in-flight iterations, total_tokens, and tool_calls in durable QueueState files allows scans to resume after server restarts without counter resets. Not checked yet OpenAI-compatible API now supports typed-part content arrays (e.g. [{"type":"text","text":"..."}]) in response bodies. Not checked yet
3 details in Everything
New XALGORIX_ITERATION_DELAY env var and Web UI setting control agent pacing to manage rate limits
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.83 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.83
└──▷ USE IT
Set a 5-second pause between agent reasoning turns to manage provider rate limits via environment variable.
shell
copy
export XALGORIX_ITERATION_DELAY=5 && xalgorix scan --target <target>
Configure the iteration delay directly in the Scanner Web UI under Runtime settings.
📍 In the Scanner Web UI, go to Settings › Runtime and set Iteration Delay.
Adds the XALGORIX_ITERATION_DELAY environment variable (range 0..300s, default 0) to configure iteration delay for the agent Not checked yet Exposes the iteration delay configuration in Scanner Web UI Settings as a Runtime setting Not checked yet
2 details in Everything
xalgorix v4.6.81 adds env vars XALGORIX_LLM_MAX_INFLIGHT and XALGORIX_MAX_CONCURRENT_AGENTS for concurrency control.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.81 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.81
Adds XALGORIX_LLM_MAX_INFLIGHT config key to control maximum simultaneous LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. requests (default: 4 × MaxInstances, clamped 1..256). Not checked yet Adds XALGORIX_MAX_CONCURRENT_AGENTS config key to serialize or parallelize specialist subagents (default: 3, clamped 1..10). Not checked yet
2 details in Everything
v4.6.77 preserves model IDs for subagents and enforces proof-integrity for LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. responses
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.77 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.77
Preserves slash-separated model identifiers (e.g. zai-org/GLM-5.3) when routing through custom OpenAI-compatible endpoints or cloud gateways, enabling consistent model targeting in complex proxy setups Not checked yet Propagates cloned LLM client configuration, temperature settings, and custom resolvers directly to delegated subagents, allowing consistent behavioral parameters across distributed inference tasks Not checked yet
2 details in Everything
Adds XALGORIX_PROXY_REQUIRED env var to enforce strict proxy routing and prevent egress leaks in xalgorix
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.76 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.76
Introduces XALGORIX_PROXY_REQUIRED=true environment variable to enforce that scan traffic routes strictly through a single upstream proxy without silent fallback to direct connections. Not checked yet Adds startup validation for XALGORIX_USE_PROXY and XALGORIX_PROXY_URL, failing immediately if missing or invalid when proxy enforcement is enabled. Not checked yet Routes core HTTP tools (httpclient, sendraw, browser, proxy_request) strictly via the configured upstream proxy, failing requests if the proxy is down. Not checked yet
3 details in Everything
xalgorix v4.6.73 introduces benchmark controls, endpoint tracking, and scan-local artifacts for specialist authorization scans.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.73 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.73
Adds a pinned real-world CVE benchmark with vulnerable/fixed controls to validate recall/stability of authorization and server-side injection detections Not checked yet Launches a bounded, deterministic specialist wave targeting authorization/business logic, server-side injection, and client/source coverage Not checked yet Keeps benchmark and agent artifacts in scan-local repository tmp/ directories rather than host /tmp Not checked yet
3 details in Everything
Adds authenticated two-account BOLA BOLA Broken Object Level Authorization, an API vulnerability class where a server fails to verify that the requesting user owns or may access the object ID they supply, letting attackers read or modify other users' data. challenge support to benchmarks
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.58 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.58
Adds Auth declaration to challenges for two-account BOLA testing, wiring role A as the scan session and role B as the comparison identity. Not checked yet Introduces the bola challenge to detect Broken Object Level Authorization flaws by replaying requests across authenticated roles. Not checked yet Adds safe-bola negative control that enforces ownership checks to prevent over-reporting scanners from passing. Not checked yet
3 details in Everything
v4.6.54 changes agent phase logic to route injection signals to specific verifiers for verified findings
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.54 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.54
PHASE 6 (injection testing) now routes parameters showing a class signal directly to the matching deterministic verifier (verify_sqli, verify_ssti, verify_xss, verify_xxe, verify_csrf, or verify_oob) instead of repeatedly probing with generic scanners Not checked yet PHASE 20 (exploit verification) maps specific injection classes to their verifiers: SQLi SQLi An attack that makes an application treat untrusted input as part of a database command. It can expose or modify data by changing what the query does. →verify_sqli, reflected/DOM DOM The browser's in-memory tree representation of a web page's structure, standardized by the W3C, giving cyber tools a programmable interface to read or manipulate page content at runtime. XSS XSS Cross-site scripting, an attack that runs malicious script in another user's browser through a trusted site. It can steal data or perform actions using the user's session. →verify_xss (including POST via data=), SSTI SSTI Server-Side Template Injection, a vulnerability class where attacker-controlled input is evaluated by a server-side template engine, enabling remote code execution or data exfiltration. →verify_ssti, XXE XXE XML External Entity injection, a class of attack against XML parsers that tricks them into fetching external resources or local files, letting an attacker read sensitive data or trigger server-side request forgery. →verify_xxe, CSRF→verify_csrf, blind→verify_oob; LFI is routed straight to report_vulnerability Not checked yet Scanners are demoted to fallback status in injection testing when a signal is detected, allowing the agent to use the one-call verifier for immediate confirmation Not checked yet
3 details in Everything
xalgorix v4.6.53 adds POST-based reflected XSS XSS Cross-site scripting, an attack that runs malicious script in another user's browser through a trusted site. It can steal data or perform actions using the user's session. confirmation via verify_xss
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.53 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.53
└──▷ TRY IT
Confirm a reflected XSS payload that only triggers via a POST form submission in CI or automated scans.
copy
$ xalgorix verify_xss --url "https://target.example.com/search" --data "search=<script>alert(1)</script>"
verify_xss now confirms POST-based reflected XSS by staging a self-submitting cross-origin form; pass url and urlencoded data to trigger a top-level POST navigation. Not checked yet method defaults to POST whenever the data argument is present in verify_xss, eliminating the need to explicitly set it for form submissions. Not checked yet
2 details in Everything
xalgorix v4.6.52 adds verify_csrf deterministic verifier for CSRF CSRF Cross-site request forgery, an attack that makes a signed-in browser send an unwanted request to a site. The site may accept the request using the victim's existing session. confirmation
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.52 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.52
└──▷ HOW TO FIND IT
Confirm CSRF vulnerability on a suspected endpoint using the new verifier.
📍 xalgorix verify_csrf --url <target_url> --body <state_change_payload>
Adds verify_csrf, a new command in the deterministic verifier family that replays requests to confirm CSRF vulnerabilities and records CWE CWE Common Weakness Enumeration, a catalog of recurring software and hardware weakness types. It describes classes of mistakes, such as injection, rather than individual vulnerability instances. -352 evidence. Not checked yet The new verify_csrf command automatically declines testing on endpoints protected by Authorization headers to prevent false positives. Not checked yet
2 details in Everything
v4.6.48 introduces deterministic verification agents for SQLi SQLi An attack that makes an application treat untrusted input as part of a database command. It can expose or modify data by changing what the query does. , SSTI SSTI Server-Side Template Injection, a vulnerability class where attacker-controlled input is evaluated by a server-side template engine, enabling remote code execution or data exfiltration. , XSS XSS Cross-site scripting, an attack that runs malicious script in another user's browser through a trusted site. It can steal data or perform actions using the user's session. , and OOB OOB Out-of-band communication: a separate channel from the primary data path, used to send control signals or exfiltrate data without touching monitored traffic, giving cyber tools a covert or resilient side channel. flaws
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.48 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.48
Adds verify_sqli, verify_ssti, verify_xss, and verify_oob verifiers to perform fast deterministic confirmation of exploitation evidence Not checked yet Improves reporting speed by prioritizing strong in-band signals over out-of-band callbacks Not checked yet
2 details in Everything
xalgorix v4.6.45 adds CSRF CSRF Cross-site request forgery, an attack that makes a signed-in browser send an unwanted request to a site. The site may accept the request using the victim's existing session. detection to the black-box benchmark with 25 new challenges
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.45 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.45
The black-box benchmark now exercises CSRF detection, completing coverage of every class the finding classifier recognizes. Not checked yet Adds a csrf positive challenge: a state-changing POST /account/email whose form carries no anti-CSRF token and whose handler requires none. Not checked yet A matching safe-account negative control serves the same form with a per-session anti-CSRF token and refuses any POST whose token is missing or wrong (403). Not checked yet
3 details in Everything
Adds native Z.AI (Zhipu GLM) provider support with automatic model ID routing.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.43 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.43
└──▷ HOW TO FIND IT
Configure the tool to use Zhipu GLM for generation by adding the provider via API key.
📍 In Settings › LLM provider list, select Z.AI (standard API) and enter your API key.
Adds Z.AI (standard API) and Z.AI Coding Plan LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. providers to the first-run setup and Settings → LLM provider list, keyed by API key. Not checked yet Routes glm-* model IDs automatically to Z.AI endpoints (https://api.z.ai/api/paas/v4 and https://api.z.ai/api/coding/paas/v4). Not checked yet Accepts GLM model IDs in any case (e.g., GLM-5.3 normalized to glm-5.3) without rejection as Z.AI's catalog evolves. Not checked yet
3 details in Everything
Adds negative controls for safe-lfi, safe-cmdi, safe-ssti, and safe-idor to the internal benchmark suite
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.42 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.42
v4.6.41 adds Negative challenge support and false-positive precision scoring to the xalgorix benchmark.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.41 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.41
└──▷ USE IT
Validate that a scanner correctly suppresses findings for secure inputs by defining a negative challenge.
text
copy
challenge: name: safe-search
Negative: true
...
Review benchmark results to identify scanners that over-report by checking the precision line.
📍 Precision: 5/5 negative controls clean, 0 false positive(s)
Challenge objects now support a Negative flag for negative-control testing where no finding is expected Not checked yet Benchmark scorecard now reports Precision: N/M negative controls clean, K false positive(s) and marks offending finding IDs Not checked yet Shipped four new negative controls: safe-search, safe-redirect, safe-sqli, and safe-fetch Not checked yet
3 details in Everything
New verify_ssti command confirms server-side template injection with randomized operand probes and ledger evidence.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.35 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.35
└──▷ TRY IT
Confirm SSTI SSTI Server-Side Template Injection, a vulnerability class where attacker-controlled input is evaluated by a server-side template engine, enabling remote code execution or data exfiltration. on a specific parameter by sending baseline and template probes.
copy
$ xalgorix verify_ssti --url 'http://example.com/app' --parameter 'template'
Confirm SSTI using an existing ledger hypothesis ID to avoid redundant scanning.
copy
$ xalgorix verify_ssti --hypothesis_id <id>
verify_ssti subcommand to confirm SSTI using randomized operands (Jinja2, Twig, Freemarker, etc.) Not checked yet
1 detail in Everything
Adds whitebox-ssti whitebox benchmark challenge for server-side template injection detection.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.34 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.34
└──▷ USE IT
Exercise the source-to-runtime bridge for SSTI SSTI Server-Side Template Injection, a vulnerability class where attacker-controlled input is evaluated by a server-side template engine, enabling remote code execution or data exfiltration. vulnerabilities in a whitebox setting using the new benchmark challenge.
shell
copy
./xalgorix benchmark run --challenge whitebox-ssti
New verify_sqli command confirms error-based SQL injection SQL injection An attack that makes an application treat untrusted input as part of a database command. It can expose or modify data by changing what the query does. in a single deterministic call.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.32 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.32
Adds verify_sqli subcommand to confirm error-based SQLi SQLi An attack that makes an application treat untrusted input as part of a database command. It can expose or modify data by changing what the query does. using DBMS error detection across scope-gated baseline/broken/balanced requests Not checked yet Introduces hypothesis_id and url parameters for verify_sqli invocation Not checked yet Shares DBMS-error detection via internal reporting.LooksLikeSQLError function as single source of truth Not checked yet
3 details in Everything
Adds whitebox-sqli benchmark challenge for SQL injection SQL injection An attack that makes an application treat untrusted input as part of a database command. It can expose or modify data by changing what the query does. source-to-runtime validation
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.30 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.30
v4.6.26 auto-seeds scan ledger from whitebox source at iteration 1 for faster exploit paths
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.26 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.26
Auto-seeds the scan ledger with dangerous sinks, reachable HTTP routes, and route-to-sink correlations as soon as a whitebox source is attached Not checked yet Enables probe_hypothesis specialists to receive concrete, source-derived leads starting from iteration 1 Not checked yet
2 details in Everything
New scan_source_routes command to map source sinks to reachable HTTP paths
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.24 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.24
xalgorix v4.6.23 adds the scan_source_sinks subcommand to automatically link source-level sinks to runtime hypotheses.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.23 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.23
└──▷ TRY IT
Connect whitebox source findings to the runtime hypothesis ledger to prioritize exploitation paths.
copy
$ xalgorix scan_source_sinks
Adds scan_source_sinks subcommand to sweep the source tree for dangerous sinks (RCE RCE Remote code execution, the ability to run attacker-chosen code on another system through a vulnerability. Its impact depends on the permissions and access that code receives. , SQLi SQLi An attack that makes an application treat untrusted input as part of a database command. It can expose or modify data by changing what the query does. , SSRF SSRF Server-side request forgery, an attack that tricks a server into making requests chosen by an attacker. Those requests can reach internal services the attacker cannot access directly. , LFI, SSTI SSTI Server-Side Template Injection, a vulnerability class where attacker-controlled input is evaluated by a server-side template engine, enabling remote code execution or data exfiltration. , deserialization deserialization Turning stored or transmitted data back into objects a program can use. Unsafe handling of untrusted data can allow unexpected behavior or code execution. , open redirect) and seed them into the shared ledger as source→sink hypotheses Not checked yet
1 detail in Everything
xalgorix v4.6.21 adds class-based benchmark scoring and per-challenge -timeout flag to xalgorix-bench
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.21 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.21
└──▷ TRY IT
Limit each challenge scan to 5 minutes to prevent stuck runs during benchmarking.
copy
$ xalgorix-bench -timeout 5m
Adds -timeout flag to xalgorix-bench to bound each challenge scan (default 8m). Not checked yet Adopts class-based benchmark scoring: a finding of the expected vulnerability class counts as solved, replacing the previous exact-endpoint requirement. Not checked yet
2 details in Everything
xalgorix v4.6.20 adds four new benchmark challenge classes to the xalgorix-bench harness
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.20 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.20
xalgorix-bench now measures eight classes, including SSRF SSRF Server-side request forgery, an attack that tricks a server into making requests chosen by an attacker. Those requests can reach internal services the attacker cannot access directly. , SSTI SSTI Server-Side Template Injection, a vulnerability class where attacker-controlled input is evaluated by a server-side template engine, enabling remote code execution or data exfiltration. , LFI, and command injection command injection An attack that makes an application execute operating-system commands supplied through untrusted input. It often happens when input is passed unsafely to a shell. Not checked yet Benchmark challenges simulate dangerous outcomes (e.g. {{7*7}} returns 49) rather than performing real fetches/executions Not checked yet
2 details in Everything
New xalgorix-bench command for measuring detection accuracy per vulnerability class using a local benchmark harness.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.19 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.19
New xalgorix-bench operator command that runs deliberately vulnerable challenge apps and prints a per-class scorecard Not checked yet
1 detail in Everything
Adds claim_next_hypothesis command for deterministic hypothesis queueing
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.17 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.17
└──▷ USE IT
Assign the top-scoring untested vulnerability hypothesis to the current agent for testing.
shell
copy
xalgorix claim_next_hypothesis --scope vuln_class
claim_next_hypothesis atomically claims the highest-confidence queued hypothesis and assigns it to the calling agent Not checked yet
1 detail in Everything
xalgorix v4.6.14 seeds uploaded context artifacts directly into the hypothesis ledger as role-scoped authorization hypotheses
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.14 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.14
Adds authz_matrix population from uploaded OpenAPI OpenAPI A vendor-neutral specification for describing REST APIs in a machine-readable format, maintained by the OpenAPI Initiative. Cyber tools use it to auto-generate clients, documentation, and validation from a single source of truth. /Swagger, HAR HAR An HTTP Archive file, a JSON-format log of all network requests and responses captured by a browser's developer tools, used by cyber tools to replay, inspect, or analyze web traffic without re-running a live session. , Postman collection, or Burp export contexts by seeding the hypothesis ledger with bounded, role-scoped authorization hypotheses Not checked yet
1 detail in Everything
xalgorix v4.6.13 enables authz_matrix to use ingested HAR HAR An HTTP Archive file, a JSON-format log of all network requests and responses captured by a browser's developer tools, used by cyber tools to replay, inspect, or analyze web traffic without re-running a live session. sessions as Role A for authenticated access-control testing
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.13 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.13
└──▷ USE IT
Test if an attacker can escalate privileges by replaying a recorded session in authz_matrix.
shell
copy
xalgorix scan --input har-session.har --authz-matrix
xalgorix v4.6.8 adds scan-scoped parallel specialists coordination via a durable hypothesis/evidence ledger and new deep-testing tools: authz_matrix and verify_xss.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.8 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.8
Adds scan-scoped parallel specialists capability coordinated by a durable hypothesis/evidence ledger for multi-agent assessments. Not checked yet Adds support for deep-testing the authz_matrix class via a new multi-role authorization matrix assessment. Not checked yet Adds browser-backed XSS XSS Cross-site scripting, an attack that runs malicious script in another user's browser through a trusted site. It can steal data or perform actions using the user's session. execution verification (verify_xss) to address weaknesses in autonomous scanners. Not checked yet
3 details in Everything
Adds XALGORIX_NOTIFY_SCAN_COMPLETE environment variable to gate scan completion notifications for Discord and Telegram.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.6.5 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.6.5
Gates all queue-level and report-level completion summaries while preserving per-vulnerability alerts Not checked yet
1 detail in Everything
xalgorix v4.5.156 adds light theme toggle and multi-file Postman upload with auth/variable resolution
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.156 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.156
└──▷ HOW TO FIND IT
Switch the UI to light mode or follow system preference directly from the web client.
📍 Use the light/dark/system toggle in the web UI theme settings.
Upload multiple Postman collection files to resolve shared variables and auth across them automatically.
📍 Upload multiple Postman files via the context upload interface.
feat(webui) : add light theme with a light/dark/system toggle to the web interface Not checked yet feat(context) : enable multi-file Postman upload with automatic variable and authentication resolution Not checked yet
2 details in Everything
xalgorix v4.5.141 adds opt-in loopback pprof debug server via XALGORIX_PPROF_ADDR
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.141 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.141
xalgorix v4.5.125 adds split-APK support and configurable OOB OOB Out-of-band communication: a separate channel from the primary data path, used to send control signals or exfiltrate data without touching monitored traffic, giving cyber tools a covert or resilient side channel. callback selection
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.125 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.125
Support split-APK bundles (.apks/.xapk/.aab) and stop rejecting sparse APKs (#384, #385) Not checked yet Choose which OOB interaction types (DNS/HTTP/SMTP SMTP The application-layer protocol used to send email between servers and from clients to servers; cyber tools use it to relay alerts, phishing simulations, or exfiltrated data over standard mail infrastructure. ) count as callbacks (#381) Not checked yet
2 details in Everything
xalgorix v4.5.121 adds UI controls to pause, resume, and stop active or pending scan instances.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.121 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.121
└──▷ HOW TO FIND IT
Pause a running scan to free up resources or investigate an issue without aborting.
📍 In the console, select the active scan instance and click Pause .
Resume a previously paused scan to continue where it left off.
📍 In the console, select the paused scan instance and click Resume .
Stop a pending scan from the admission queue to clear backlog.
📍 In the console, select the pending scan instance and click Stop .
v4.5.117 preserves iteration counts across restarts and injects Resume Context Briefing for LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. agents.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.117 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.117
Preserves cumulative iteration count across server restarts to maintain scanning context Not checked yet Injects a structured Resume Context Briefing into the LLM context for continuous agent testing Not checked yet
2 details in Everything
xalgorix v4.5.114 adds auto-inference for verification_method and exploitation_proof in report_vulnerability
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.114 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.114
report_vulnerability now automatically extracts verification_method and exploitation_proof from the description/technical analysis when omitted, preventing call rejection. Not checked yet Ensures all verified findings land on the dashboard without requiring explicit LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. arguments for proof fields. Not checked yet
2 details in Everything
xalgorix v4.5.111 adds HTTP redirect tracking and repeat loop exemptions
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.111 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.111
Adds hookRedirectDetector to automatically detect HTTP 301/302/307/308 redirects and prompt the agent to use curl -L and target the destination path directly. Not checked yet Exempts knowledge tools (read_skill, list_skills, search_skills) and tool parameter validation error outputs from result repeat loop counters. Not checked yet
2 details in Everything
xalgorix v4.5.100 adds Target Health & IP Ban Monitor hook for mid-scan network failure detection
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.100 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.100
v4.5.98 adds Recursive JS Sub-Chunk Tree Traversal, Automated Parameter Mining, HTTP Verb Tampering, and Header Override Permutations for new attack surface attack surface The exposed systems, interfaces and access points an attacker could try to use. It includes internet-facing services as well as identities, software and other reachable resources. discovery.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.98 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.98
Adds Recursive JS Sub-Chunk Tree Traversal to find hidden endpoints in Single Page Apps Not checked yet Adds Automated Parameter Mining to discover hidden parameters before injection testing Not checked yet Adds HTTP Verb Tampering and Header Override Permutations for route and authentication bypasses Not checked yet
3 details in Everything
v4.5.95 adds mandatory OAST OAST Out-of-band Application Security Testing: a technique for detecting vulnerabilities by having a target system make DNS or HTTP callbacks to an attacker-controlled server, revealing flaws that produce no visible in-band response. gatekeeper check requiring at least one callback probe for blind vulnerability classes
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.95 https://github.com/xalgord/xalgorix.git
$ git checkout v4.5.95
xalgorix v4.5.77 adds custom identifying headers for scan traffic via -H flag.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.77 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.77
Registers arjun, uro, and x8 tools for auto-installation in the terminal subsystem Not checked yet
1 detail in Everything
New commands to list and delete provider profiles in Settings and a structural task planner for the agent.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.73 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.73
Adds configurable day and time of day for scan schedules via web UI
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.72 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.72
Adds opt-in local-target scanning for self-hosted xalgorix installs and multi-line scan instructions
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.62 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.62
Enables scopeguard to opt-in local-target scanning for self-hosted installations Not checked yet Supports multi-line scan instructions in the web UI Not checked yet
2 details in Everything
xalgorix v4.5.57 adds the search_skills tool for concept-based skill retrieval
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.57 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.57
Adds Star on GitHub CTA to dashboard and persists Settings via Docker compose.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.52 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.52
└──▷ USE IT
Run xalgorix with settings persistence using the new compose file.
shell
copy
docker compose up -d
# Settings are persisted to /data on the host volume mounted in compose
Adds docker-compose file and persisting /data volume for Settings Not checked yet Adds Star on GitHub call-to-action in the dashboard sidebar Not checked yet
2 details in Everything
xalgorix v4.5.33 adds scan-summary feature, independent finding verification with false-positive hardening
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.33 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.33
eb7d82e introduces the scan-summary feature alongside LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. resolver/client updates and a web UI refresh (PR #155). Not checked yet Independent finding verification is now available, providing false-positive hardening for findings workflows (PR #154). Not checked yet
2 details in Everything
xalgorix v4.5.11 adds per-role LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. temperatures and surface-area-aware finish gating.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.11 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.11
Adds per-role LLM_TEMPERATURE_SCANNER=0.2, LLM_TEMPERATURE_VALIDATOR=0.0, LLM_TEMPERATURE_REPORTER=0.3 environment variables to control model stochasticity. Not checked yet
1 detail in Everything
xalgorix v4.5.10 adds per-role LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. temperature control and surface-area-aware finish gating with depth tracking
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.10 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.10
per-role LLM temperature allows setting distinct temperatures for scanner (0.2), validator (0.0), and reporter (0.3) roles. Not checked yet
1 detail in Everything
xalgorix v4.5.3 adds --restart-when-idle flag and improves web scan list pagination performance
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.3 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.3
└──▷ TRY IT
Automatically restart the server when idle to free resources or apply config changes without manual intervention.
copy
$ xalgorix server --restart-when-idle
xalgorix v4.5.1 adds shorthand aliases for all 673 skills and MITRE MITRE A US federally funded research and development center that maintains public frameworks—ATT&CK, CVE, D3FEND—that cyber tools use to classify threats, vulnerabilities, and defensive techniques in a common language. mapping in misspell linter
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.1 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.1
Adds shorthand alias for every skill (673 new) in skills configuration block with automatic collision checking Not checked yet Enables MITRE technique references in the misspell linter subsystem Not checked yet
2 details in Everything
v4.5.0 adds server-side pagination, search, and filters to the web UI for scans and instances.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.5.0 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.5.0
Enriches all 858 skills with type-aware validation/confirmation sections Not checked yet
1 detail in Everything
Introduces RAM RAM Working memory that holds programs and data while a computer is running. Its contents usually disappear when power is removed. -only scan admission by removing dead CPU CPU The main general-purpose processor in a computer. It runs application instructions and coordinates work with memory and other hardware. /disk slot inputs.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.4.20 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.4.20
Xalgorix v4.4.16 adds CheckRead with a built-in deny-list and XALGORIX_READ_DENY_LIST env var for filesystem reads.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.4.16 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.4.16
└──▷ TRY IT
Allow scanning additional sensitive directories (e.g., custom secret stores) alongside the defaults without modifying source code.
copy
$ export XALGORIX_READ_DENY_LIST="/home/user/secret:/opt/private/data"
Check which paths are currently blocked for reads via the API to verify policy configuration.
copy
$ curl -s /api/status | jq '.read_deny'
Use the fileedit command to read system wordlists that were previously rejected by the sandbox.
copy
$ xalgorix fileedit view /usr/share/seclists/rockyou.txt
Adds CheckRead function to allow reads outside the workspace by default while blocking sensitive paths via a built-in deny-list Not checked yet Adds XALGORIX_READ_DENY_LIST environment variable (colon-separated) to extend the default deny-list for filesystem tools Not checked yet Exposes the active deny-list in the read_deny field on the /api/status endpoint Not checked yet
3 details in Everything
v4.4.15 adds findings pagination, deduplication, and stable summary API; relocates data dir.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.4.15 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.4.15
findings_consistency_and_pagination spec enables pagination on the Findings page with configurable size in [25, 50, 100, 200] rows (default 50) and URL-synced params ?page=&size=. Not checked yet New /api/findings/summary endpoint provides stable, deduplicated counts with ETag ETag An HTTP response header that carries an opaque version token for a resource, letting clients skip re-downloading unchanged content by sending the token back in a conditional request. /304 support for consistent totals during background polling. Not checked yet effectiveVulnCount(inst, sess) consolidates vulnerability counting into a single monotonic source that reads in-memory and persists without dropping. Not checked yet reporting.PromoteToParent called on every report_vulnerability ensures child scan findings immediately aggregate to the parent record even during panics. Not checked yet Workspace default migrated to ~/.xalgorix/data with automatic non-destructive import from legacy ~/xalgorix-data/ directory via sentinel file .legacy-imported. Not checked yet
+2 more
show less
safe.Recover panic boundaries added for child findings and HTTP middleware; counters exposed (PanicsRecovered, PathRejections, WatchdogKills, AdmissionRefusals). Not checked yet sandbox.Policy path allow-list checks enforced for every fs-touching tool to enhance workspace isolation. Not checked yet
7 details in Everything
└──▷ BREAKING ON UPGRADE
! Workspace data directory default changed from ~/xalgorix-data/ to ~/.xalgorix/data; legacy path preserved but marked for removal.
xalgorix v4.4.8 adds recurring scheduled scans with disk-persisted configs and a frontend dashboard for management.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.4.8 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.4.8
└──▷ HOW TO FIND IT
Manage scan schedules directly through the new UI interface.
📍 Open the frontend dashboard › Use CRUD modal to create/edit a schedule
Add file_upload vuln type to all mapping tables Not checked yet cweIDToName reverse lookup and owaspIDToName direct lookup maps added for security framework mappings (CWE CWE Common Weakness Enumeration, a catalog of recurring software and hardware weakness types. It describes classes of mistakes, such as injection, rather than individual vulnerability instances. , OWASP OWASP A nonprofit foundation that publishes open standards, guides, and ranked vulnerability lists—most famously the OWASP Top 10—giving cyber tools a vendor-neutral benchmark to test or report against. Top 10, PTES) Not checked yet Frontend dashboard with CRUD CRUD The four basic database operations — Create, Read, Update, Delete — used as shorthand for any system that manages persistent data records, giving cyber tools a standard model for interacting with stored information. modal, status toggles, and manual triggers for managing scans Not checked yet Recurring scheduled scans (hourly/daily/weekly/monthly) with disk-persisted configs Not checked yet
4 details in Everything
xalgorix v4.4.3 adds DELETE /api/scans/{id}/vulns/{vulnId} API and bulk findings UI actions
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.4.3 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.4.3
DELETE /api/scans/{id}/vulns/{vulnId} endpoint to delete individual vulnerabilities from a scan Not checked yet Findings page checkbox selection, bulk delete, and per-row action menu Not checked yet Scan detail FindingsTab with checkbox + bulk delete UI Not checked yet
3 details in Everything
xalgorix v4.4.2 adds live feed event export in JSON, JSONL JSONL A text format where each line is a self-contained, valid JSON object, making it easy to stream, append, and process large datasets line by line without parsing a full JSON array into memory. , and transcript formats.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.4.2 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.4.2
xalgorix v4.4.0 adds a new React Web UI embedded in release binaries and settings-backed env configuration.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.4.0 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.4.0
Ships the new React Web UI as embedded static assets for Go installs and release binaries Not checked yet Adds settings-backed environment configuration support Not checked yet
2 details in Everything
xalgorix v4.3.12 introduces persisted scan targets and extended ScanRecord fields for configuration flexibility
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.3.12 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.3.12
└──▷ USE IT
Configure a scan with specific metadata and severity filtering using the new ScanRecord fields.
json
copy
{
"Name": "critical-scan",
"Instruction": "Check for CVEs in production",
"SeverityFilter": "high",
"DiscordWebhook": "https://discord.com/api/webhooks/..."
}
ScanRecord struct gains Name , Instruction , SeverityFilter, and DiscordWebhook fields to control scan metadata and notifications Not checked yet Saved scan targets are now persisted to disk, allowing them to survive server restarts Not checked yet
2 details in Everything
xalgorix v4.3.8 adds logo upload endpoint, Discord severity threshold, local IP scan prevention, and phase label sync
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.3.8 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.3.8
Adds /api/upload-logo endpoint for logo uploads with image validation and live preview UI support Not checked yet Enables XALGORIX_DISCORD_MIN_SEVERITY environment variable to set the minimum severity threshold for Discord notifications Not checked yet Syncs methodology phase labels with the backend's 22-phase map for consistent reporting Not checked yet
3 details in Everything
xalgorix v4.3.7 adds proxy support (HTTP/HTTPS HTTPS HTTP carried over an encrypted TLS connection. It protects traffic in transit and lets a browser verify the server's certificate. /SOCKS5 SOCKS5 A proxy protocol that relays TCP and UDP traffic through an intermediary server without inspecting it, letting cyber tools tunnel connections through firewalls or pivot hosts with minimal overhead. ) with auto-rotation
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.3.7 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.3.7
Adds secure session destruction endpoint POST /api/auth/logout
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.2.7 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.2.7
└──▷ TRY IT
Terminate active sessions via API for automated security cleanup or audit trails.
copy
$ curl -X POST /api/auth/logout
Adds cyberpunk-styled logout button with neon hover effects to dashboard header Not checked yet
1 detail in Everything
xalgorix v4.2.5 adds natural shorthand aliases for LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. agent skills like read_skill(name="sql-injection") .
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.2.5 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.2.5
└──▷ USE IT
Scan for SQL injection SQL injection An attack that makes an application treat untrusted input as part of a database command. It can expose or modify data by changing what the query does. using a short, memorable alias instead of the full skill name.
code
copy
read_skill(name="sql-injection")
Adds support for ~100 natural shorthand aliases (e.g., sql-injection, xss, k8s-privesc) in the read_skill function to map to existing verbose skill directory names Not checked yet
1 detail in Everything
xalgorix v4.2.3 adds security hardening, observability helpers, and dynamic skill discovery
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.2.3 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.2.3
logRecover() helper replaces silent recover() calls with labeled panic + stack trace logging for improved visibility. Not checked yet Dynamic skill category discovery allows embedded markdown skills to be organized without hardcoded lists. Not checked yet Auto-chmod of .env files to 0600 tightens file permissions on deployment. Not checked yet
3 details in Everything
xalgorix v4.2.0 adds Phase 21 behavioral fuzzing and new zero-day zero-day A vulnerability for which defenders have had no prior opportunity to apply a fix, often because it was previously unknown. A zero-day exploit takes advantage of that gap. hunting skills
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.2.0 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.2.0
Introduces Phase 21 to the execution methodology, enabling behavioral fuzzing, parser differential testing, timing side-channels, type confusion, and state machine attacks Not checked yet Adds zero-day-hunting and response-anomaly-detection as new actionable skills that auto-load when relevant keywords are detected in instructions Not checked yet
2 details in Everything
v4.1.17 expands single-site scan scope to include sibling subdomains and root domain
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.1.17 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.1.17
Automatically follows auth/SSO SSO Single sign-on, using one identity provider to sign in to multiple applications. It reduces separate passwords and centralizes account and access management. /API redirects to sibling subdomains Not checked yet
1 detail in Everything
xalgorix v4.1.15 adds auto-downloading Chromium, custom browser path via XALGORIX_BROWSER_PATH, and the pageagent tool for script injection.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.1.15 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.1.15
Adds auto-download of embedded Chromium binary (~170MB on first run, cached) Not checked yet Adds XALGORIX_BROWSER_PATH environment variable to override the browser executable path Not checked yet
2 details in Everything
Adds IsResume flag to control auto-resume state preservation and orphan process cleanup in xalgorix v4.1.13.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.1.13 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.1.13
Introduces IsResume flag for proper auto-resume detection. Not checked yet Preserves vulns, notes, and recon files on auto-resume instead of resetting state. Not checked yet
2 details in Everything
v4.1.0 adds Clickjacking, SSTI SSTI Server-Side Template Injection, a vulnerability class where attacker-controlled input is evaluated by a server-side template engine, enabling remote code execution or data exfiltration. , and Web LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. attack skills to xalgorix cyber tooling
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.1.0 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.1.0
Adds Clickjacking vulnerability skill with detection and exploitation techniques. Not checked yet Adds SSTI (Server-Side Template Injection) skill covering Jinja2, Twig, Freemarker, etc. Not checked yet Adds Web LLM Attacks skill for prompt injection prompt injection An attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools. and data exfiltration data exfiltration Moving data out of a system or organization without authorization. It can happen through network transfers, removable media or other channels. via AI integrations. Not checked yet
3 details in Everything
Adds delete scan functionality and infinite scan timeout; removes wildcard scan target limit.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.0.12 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.0.12
LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. Settings UI upgrade adds model and API base dropdowns plus explicit API base support.
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.0.4 https://github.com/xalgorix/xalgorix.git
$ git checkout v4.0.4
└──▷ HOW TO FIND IT
Select a specific provider model directly from the LLM Settings dropdown instead of manual typing.
📍 Open the LLM Settings UI and select the desired Model from the dropdown menu.
Configure MiniMax to use its China-specific endpoint via the API Base dropdown.
📍 Open the LLM Settings UI and select api.minimax.cn from the API Base dropdown.
Adds Model dropdown to LLM Settings UI that populates current models per provider Not checked yet Adds API Base dropdown with provider-specific options including MiniMax Global (api.minimax.io) and China (api.minimax.cn) endpoints Not checked yet Backend resolveEndpoint() now respects explicit API base provided via the web UI Not checked yet
3 details in Everything
xalgorix v4.0.0 adds real-time vulnerability reporting and multi-step login support
└──▷ GET THIS VERSION
copy
$ git clone --branch v4.0.0 https://github.com/xalgord/xalgorix.git
$ git checkout v4.0.0
└──▷ USE IT
Install the new major version of the cyber tooling.
shell
copy
go install github.com/xalgord/xalgorix/v4/cmd/xalgorix@latest
Enables real-time vulnerability reporting instead of bulk processing. Not checked yet Adds multi-step login support including SSO SSO Single sign-on, using one identity provider to sign in to multiple applications. It reduces separate passwords and centralizes account and access management. , OAuth, 2FA 2FA Two-Factor Authentication, an authentication scheme requiring a second proof of identity beyond a password, so a stolen credential alone cannot grant access. , and magic links. Not checked yet Requires authentication for IDOR IDOR Insecure Direct Object Reference, a web vulnerability where an application exposes internal identifiers (IDs, filenames) without access checks, letting an attacker substitute another user's ID to reach their data. testing. Not checked yet
3 details in Everything
└──▷ BREAKING ON UPGRADE
! Module path changed from github.com/xalgord/xalgorix/v3 to github.com/xalgord/xalgorix/v4! Update install command: go install github.com/xalgord/xalgorix/v4/cmd/xalgorix@latest
Adds pending request queuing to the dashboard multi-instance workflow
└──▷ GET THIS VERSION
copy
$ git clone --branch v3.19.10 https://github.com/xalgord/xalgorix.git
$ git checkout v3.19.10
Adds pending state to the web server for concurrent scan requests in the dashboard Not checked yet Enables sequential queueing of distinct dashboard Instance Cards with a Pending status badge Not checked yet
2 details in Everything
xalgorix v3.19.0 adds on-demand skill loading with new detection capabilities for WebSocket WebSocket A protocol that upgrades an HTTP connection into a persistent, full-duplex channel, letting a server push data to a client without repeated polling. Cyber tools use it for real-time event streaming and live command output. Hijacking and Insecure Deserialization Deserialization Turning stored or transmitted data back into objects a program can use. Unsafe handling of untrusted data can allow unexpected behavior or code execution. .
└──▷ GET THIS VERSION
copy
$ git clone --branch v3.19.0 https://github.com/xalgorix/xalgorix.git
$ git checkout v3.19.0
read_skill and list_skills tools allow agents to dynamically load deep vulnerability knowledge during scans Not checked yet New high-impact skills: WebSocket Hijacking, 2FA 2FA Two-Factor Authentication, an authentication scheme requiring a second proof of identity beyond a password, so a stolen credential alone cannot grant access. /MFA MFA Multi-factor authentication, requiring proof from different categories, such as a password and a hardware key, to sign in. It reduces reliance on a password alone. Bypass, Insecure Deserialization, HTTP/2 HTTP/2 The second major version of the HTTP protocol, standardized by the IETF, adding multiplexed streams, header compression, and binary framing over a single TCP connection to cut latency and reduce overhead. Attacks Not checked yet
2 details in Everything
xalgorix v3.16.0 adds dashboard auth, 13 browser commands, and HackerOne-aligned severity mapping
└──▷ GET THIS VERSION
copy
$ git clone --branch v3.16.0 https://github.com/xalgorix/xalgorix.git
$ git checkout v3.16.0
Adds XALGORIX_USERNAME and XALGORIX_PASSWORD environment variables and ~/.xalgorix.env file for session-based dashboard authentication Not checked yet Introduces 13 new browser tool commands: submit, fill_form, get_cookies, set_cookie, save_session, load_session, wait, select, get_url, iframe, main_frame, and extract_links Not checked yet Adds automatic CVSS CVSS Common Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. 3.1 range enforcement and severity caps aligned with HackerOne guidelines Not checked yet
3 details in Everything
v3.10.8 adds per-command timeouts and a watchdog to prevent infinite scan hangs in xalgorix
└──▷ GET THIS VERSION
copy
$ git clone --branch v3.10.8 https://github.com/xalgorix/xalgorix.git
$ git checkout v3.10.8
Adds implicit per-command timeouts: 10 minutes for lightweight commands (curl, wget), 30 minutes for heavy tools (nmap, nuclei, ffuf, gobuster, sqlmap, masscan), and a hard 2-hour ceiling for any command. Not checked yet Introduces a scan-level timeout that force-stops an entire scan after 2 hours of total execution time. Not checked yet
2 details in Everything
Xalgorix v3.10.6 adds automatic self-updating via GitHub releases on startup.
└──▷ GET THIS VERSION
copy
$ git clone --branch v3.10.6 https://github.com/xalgorix/xalgorix.git
$ git checkout v3.10.6
└──▷ TRY IT
Run xalgorix with explicit manual update request using the new flag.
Adds automatic binary update check and replacement on startup by querying the GitHub Releases API. Not checked yet Supports --update flag to control manual update mode behavior. Not checked yet
2 details in Everything
Xalgorix v1.0.0 Stable Release adds DAST DAST Dynamic application security testing, probing a running application from the outside for vulnerabilities. It checks behavior exposed through the application's interfaces. /Wildcard modes and 3-phase scanning.
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.0.0 https://github.com/xalgorix/xalgorix.git
$ git checkout v1.0.0
Supports Single, DAST, and Wildcard scan modes for automated vulnerability discovery Not checked yet Implements 3-phase scanning workflow: Recon → Vulnerability Testing → Manual Exploitation Not checked yet Enables Gemini web search integration to enhance reconnaissance Not checked yet Configures stricter vulnerability reporting that filters out non-exploitable findings Not checked yet Mandates WAF WAF A reverse proxy that inspects HTTP/HTTPS traffic and blocks requests matching known attack patterns such as SQL injection or XSS, sitting between clients and a web application to filter malicious input before it reaches the server. bypass capabilities during scanning operations Not checked yet
+2 more
show less
Introduces tool auto-installation with dynamic PATH configuration Not checked yet Adds dynamic home directory support for flexible deployment Not checked yet
7 details in Everything
v0.6.0 adds severity filtering, out-of-scope fields, and Caido browser integration via CAIDO_PORT
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.6.0 https://github.com/xalgorix/xalgorix.git
$ git checkout v0.6.0
Adds CAIDO_PORT environment variable to configure the browser proxy for Playwright/Caido integration Not checked yet Introduces UI support for Severity Filter (Critical/High/Medium/Low/Info) to control result visibility Not checked yet Adds Out of Scope field to tag and exclude specific assets from analysis Not checked yet
3 details in Everything
xalgorix v0.5.0 adds severity filtering, service management commands, and automated Discord PDF reports
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.5.0 https://github.com/xalgorix/xalgorix.git
$ git checkout v0.5.0
Introduces a Severity Filter to filter vulnerabilities by severity (Critical/High/Medium/Low/Info) Not checked yet Enables automatic PDF report delivery to a configured Discord channel upon scan completion Not checked yet
2 details in Everything
xalgorix v0.4.0 introduces rate limiting, circuit breaker, encoding bypass detection, and scan queue persistence
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.4.0 https://github.com/xalgorix/xalgorix.git
$ git checkout v0.4.0
Adds configurable RATE_LIMITING via environment variables or Web UI to control tool execution rates Not checked yet Adds Circuit Breaker logic that auto-blocks failing tools after 5 failures with 60s recovery period Not checked yet Adds Encoding Bypass Detection for base64, hex, and URL-encoded malicious commands Not checked yet Adds Scan Queue Persistence to resume interrupted scans after server restart Not checked yet
4 details in Everything
xalgorix v0.3.0 adds env file config, daemon mode, self-update flag, and PDF export
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.3.0 https://github.com/xalgorix/xalgorix.git
$ git checkout v0.3.0
└──▷ TRY IT
Run xalgorix in the background so it serves results continuously without keeping a terminal window open.
Trigger a self-update directly from the command line to get the latest version without manual reinstallation.
Adds ~/.xalgorix.env as the new configuration file location, with support for running under sudo. Not checked yet Adds --web -d flags to enable Daemon Mode, running xalgorix as a background service. Not checked yet Adds -up flag to trigger an in-tool self-update mechanism (xalgorix -up). Not checked yet Introduces PDF pentest report auto-generation upon scan completion. Not checked yet
4 details in Everything
xalgorix v0.2.7 adds Mobile Responsive UI for better usability on small screens
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.2.7 https://github.com/xalgorix/xalgorix.git
$ git checkout v0.2.7
Adds full responsive CSS to the UI, enabling proper rendering on tablet, phone, and small screens Not checked yet
1 detail in Everything
xalgorix v0.2.4 adds PDF report generation, self-update -up flag, config file support, and daemon mode.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.2.4 https://github.com/xalgorix/xalgorix.git
$ git checkout v0.2.4
└──▷ TRY IT
Run xalgorix in the background as a daemon service.
Update xalgorix to the latest version from the command line.
Adds ~/.xalgorix.env configuration file support that works with sudo Not checked yet Adds -up CLI flag for self-updating to the latest version Not checked yet Adds --web -d flags to run xalgorix in background daemon mode Not checked yet Adds PDF report generation feature on scan completion Not checked yet
4 details in Everything