Decepticon

No entries match these filters. Clear filters

1. v1.6.0 Oct 25, 2026 · issue 120

   Decepticon v1.6.0 persists per-thread AI controls and adds compaction thresholds, SSESSEServer-Sent Events, a browser API and HTTP-based protocol where a server pushes a stream of text events to a client over a single long-lived connection, giving tools a lightweight alternative to WebSockets for one-way real-time updates. recovery, and improved checkpoint diagnostics.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v1.6.0 https://github.com/BitterSecurity/Decepticon.git
   # already have the repo? check out this version:
   $ git checkout v1.6.0

   └──▷ TRY IT
   Upgrade an existing Decepticon installation to v1.6.0, fetching the new launcher, configuration, and all eight versioned images at once.
   copy

   $ decepticon update --channel latest

   • Installs or updates to v1.6.0 via decepticon update --channel latest, pulling matching launcher, configuration, and versioned images in one step.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Persists per-thread model, fallback, reasoning effort, and compaction controls across sessions, including a verified 4096-token compaction threshold.
   • Improves scoped checkpoint diagnostics, SSE recovery, and transcript navigation across CLI, web, and server surfaces.
   • Improves localized slash-command support inside the interactive CLI.
   • Aligns web and server workspace roots, preventing stale terminal socket handoffs.

   +2 more show less

   • Preserves private Plan and Red drafts and history across container recreation.
   • Switches Kali downloads to the official Kali CDNCDNA content delivery network, servers distributed across locations that cache and serve content near readers. It can reduce loading time and traffic to the original server..
   • Installs or updates to v1.6.0 via decepticon update --channel latest, pulling matching launcher, configuration, and versioned images in one step.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   6 more in Everything

2. v1.5.2 Oct 18, 2026 · issue 113

   Decepticon v1.5.2 lets new CLI engagements auto-generate RoE and jump straight into Red by supplying a name and target URL.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v1.5.2 https://github.com/BitterSecurity/Decepticon.git
   # already have the repo? check out this version:
   $ git checkout v1.5.2

   └──▷ TRY IT
   Start a new engagement directly into Red without a pre-written scope file — supply a name and root URL and Decepticon handles RoE generation and approval automatically.
   copy

   $ decepticon update --channel latest
   decepticon

   • New CLI engagement creation prompts for an engagement name and target web URL, then automatically generates the RoE and its matching approval and opens Red for the first request — no scope JSON file, manual import/approval command, or Soundwave planning session required.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Updates via decepticon update --channel latest now include a shipped-image version check to detect mismatched CLI image metadata at install time.
   • New CLI engagement creation prompts for an engagement name and target web URL, then automatically generates the RoE and its matching approval and opens Red for the first request — no scope JSON file, manual import/approval command, or Soundwave planning session required.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   1 more in Everything

3. v1.5.0 Oct 17, 2026 · issue 112

   Decepticon v1.5.0 adds session resume, Blue monitoring, scoped Red approval, and OpenAI-compatible gateway support.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v1.5.0 https://github.com/BitterSecurity/Decepticon.git
   # already have the repo? check out this version:
   $ git checkout v1.5.0

   └──▷ TRY IT
   Resume a prior engagement mid-session without losing pending questions or queued instructions.
   copy

   $ decepticon --resume

   Import a pre-reviewed scope file and explicitly approve Red execution authority before any offensive action runs.
   copy

   $ /scope import targets.txt --confirm
   /mode red approve

   Pull structured diagnostics from a running Decepticon instance for automated health checks or CI pipelines.
   copy

   $ decepticon doctor --json

   • Adds --resume / --continue flags to restore scoped conversations and pending questions across sessions, with failed runs retaining pending input and cross-session queue isolation, plus a /queue command to manage waiting instructions between sessions.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds /scope import <file> --confirm command to import reviewed scope, requiring explicit /mode red approve before granting execution authority.
   • Adds /alerts and /status commands for Blue Cell investigation details and connectivity monitoring during Red conversations, plus an independent Blue monitoring state with compact notifications during Red conversations.
   • Adds decepticon doctor --json for read-only diagnostics output.
   • Adds conversation export to Markdown or JSON formats.

   +3 more show less

   • Adds optional vendor-neutral OpenAI-compatible gateway support.
   • Adds persistent light/dark dashboard preference.
   • Name-only engagement creation now opens Red directly without requiring a plan.
   • Adds --resume / --continue flags to restore scoped conversations and pending questions across sessions, with failed runs retaining pending input and cross-session queue isolation, plus a /queue command to manage waiting instructions between sessions.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds /scope import <file> --confirm command to import reviewed scope, requiring explicit /mode red approve before granting execution authority.
   • Adds /alerts and /status commands for Blue Cell investigation details and connectivity monitoring during Red conversations, plus an independent Blue monitoring state with compact notifications during Red conversations.
   • Adds conversation export to Markdown or JSON formats.
   • Adds optional vendor-neutral OpenAI-compatible gateway support.

   +1 more show less

   • Adds persistent light/dark dashboard preference.

   2 more in Everything

   └──▷ BREAKING ON UPGRADE

   • !The new validation and role-purpose endpoints require the updated server — launcher and Docker images/configuration must be updated together.
   • !Unscoped legacy conversations are now available only under /sessions legacy and require explicit linking to their engagement.
   • !Existing workflow checkpoints retain their pinned source; to change a workflow, a new execution conversation must be created.
4. v1.4.3 Oct 14, 2026 · issue 108

   Decepticon v1.4.3 adds mid-session mode switching via /mode plan, /mode red, and /mode blue, and streamlines engagement creation to name-only.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v1.4.3 https://github.com/BitterSecurity/Decepticon.git
   # already have the repo? check out this version:
   $ git checkout v1.4.3

   └──▷ TRY IT
   Switch from a running Plan session into Red mid-engagement once the plan is validated and scope is approved.
   copy

   $ /mode red approve

   Jump to Blue investigation while Red is running, then return to Plan — each thread resumes where it left off.
   copy

   $ /mode blue
   # ... investigate sensor events ...
   /mode plan

   Recover a paused question that was interrupted to run a command during a session.
   copy

   $ /resume

   • Adds /mode plan, /mode red, and /mode blue commands to switch conversations during a running session, preserving each mode's thread, pending question, and paused checkpoint; failure to pause keeps the original mode.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds /mode red approve to grant execution permission for Red operations against the validated Plan scope.
   • Adds /resume command to reopen a paused question after escaping with Escape to enter commands.
   • Simplifies engagement creation to require only a name, opening directly into Plan without blocking on target, mode, or authorization-reference inputs.
   • Adds /mode plan, /mode red, and /mode blue commands to switch conversations during a running session, preserving each mode's thread, pending question, and paused checkpoint; failure to pause keeps the original mode.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds /mode red approve to grant execution permission for Red operations against the validated Plan scope.
   • Adds /resume command to reopen a paused question after escaping with Escape to enter commands.

   1 more in Everything

5. v1.4.0 Oct 12, 2026 · issue 105

   Decepticon v1.4.0 adds Plan, Red, and Blue run modes to the CLI /mode command, launcher, and web dashboard, plus selectable local Ollama model IDs.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v1.4.0 https://github.com/BitterSecurity/Decepticon.git
   # already have the repo? check out this version:
   $ git checkout v1.4.0

   └──▷ TRY IT
   Switch the running session into Red mode to begin an engagement directly from a confirmed target without going through the full Plan interview.
   copy

   $ /mode Red

   Switch into Plan mode to invoke Soundwave's eight-document planning interview before committing to a Red engagement.
   copy

   $ /mode Plan

   Switch into Blue mode to bring the resident defensive agent into the explicit mode flow, triggering sensor and monitor health checks alongside the engagement.
   copy

   $ /mode Blue

   • Adds Plan, Red, and Blue run modes switchable via the CLI /mode command, OSSOSSOpen-source software: code whose licence permits anyone to inspect, modify, and redistribute it. Cyber tools build on it to reuse auditable libraries without licensing fees. launcher, and web live view; Red mode requires one confirmed target and authorization reference, and rechecks RoE approval on start; Blue mode checks both sensor and monitor health.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Brings current production LiteLLM model routes into OSS agents and CLI.
   • Adds five additional Plan document views, making all nine planning documents visible in the Plan view.
   • Adds engagement-scoped report tool calls.
   • Adds complete LiteLLM model count and concise service health details to the health surface.
   • Adds Plan, Red, and Blue run modes switchable via the CLI /mode command, OSSOSSOpen-source software: code whose licence permits anyone to inspect, modify, and redistribute it. Cyber tools build on it to reuse auditable libraries without licensing fees. launcher, and web live view; Red mode requires one confirmed target and authorization reference, and rechecks RoE approval on start; Blue mode checks both sensor and monitor health.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds engagement-scoped report tool calls.

   3 more in Everything

6. v1.1.44 Sep 19, 2026 · issue 080

   Decepticon v1.1.44 adds Autohunt autonomous planning, AWS Bedrock and Kimi provider routes, and new evidence controls including severity ceilings and streaming.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v1.1.44 https://github.com/BitterSecurity/Decepticon.git
   # already have the repo? check out this version:
   $ git checkout v1.1.44

   └──▷ TRY IT
   Disable telemetry at install time so no usage data is shared, without relying on the onboarding wizard prompt.
   copy

   $ DECEPTICON_TELEMETRY=off decepticon onboard

   Persist the telemetry opt-out across all future runs by setting it in the environment file the launcher reads.
   .env copy

   DECEPTICON_TELEMETRY=off

   • Adds DECEPTICON_TELEMETRY=off environment variable to disable telemetry sharing without going through the onboarding wizard.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds Autohunt autonomous planning lane that accepts an explicitly authorized target and validates planning documents and scope before marking an engagement ready, separate from the default Soundwave interview-first planner.
   • Adds native AWS Bedrock default routes and Kimi for Coding provider support, including launcher onboarding.
   • Adds executable specialist validation contracts covering capability contracts, evidence validation, lane scorecards, held-out evaluation fixtures, CVE-Bench dispatch plumbing, and pinned MITREMITREA 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. STIXSTIXStructured Threat Information Expression, a standard format for describing threats, indicators and their relationships. It helps organizations exchange threat intelligence in a usable form. bootstrapping.
   • Adds local API ingestion, dependency reachability evidence, and severity ceilings to evidence controls; provider calls gain extra-header and streaming controls; dynamic workload teardown is scoped by engagement and run.

   +1 more show less

   • Adds typed finding-location guidance in the skill catalog.
   • Adds DECEPTICON_TELEMETRY=off environment variable to disable telemetry sharing without going through the onboarding wizard.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds Autohunt autonomous planning lane that accepts an explicitly authorized target and validates planning documents and scope before marking an engagement ready, separate from the default Soundwave interview-first planner.
   • Adds native AWS Bedrock default routes and Kimi for Coding provider support, including launcher onboarding.
   • Adds executable specialist validation contracts covering capability contracts, evidence validation, lane scorecards, held-out evaluation fixtures, CVE-Bench dispatch plumbing, and pinned MITREMITREA 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. STIXSTIXStructured Threat Information Expression, a standard format for describing threats, indicators and their relationships. It helps organizations exchange threat intelligence in a usable form. bootstrapping.

   2 more in Everything

   └──▷ BREAKING ON UPGRADE

   • !Existing installations using default launcher credentials or missing secrets must update their configuration before starting — the release now rejects default credentials on startup.
7. v1.1.29 Jul 2, 2026 · release notes

   Adds Anthropic prompt caching support to the Claude Code OAuth LiteLLM path

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v1.1.29 https://github.com/PurpleAILAB/Decepticon.git
   # already have the repo? check out this version:
   $ git checkout v1.1.29

   • Surfaces cache_creation_input_tokens and cache_read_input_tokens in usage output for non-streaming and streaming responses.
   • Includes cache token metrics in prompt_tokens so LiteLLM prices cache reads at 0.1x and writes at 1.25x.
   • Adds cache_control breakpoint to the last message block to cache system prompt, tool results, and history across turns.

   3 details in Everything

   └──▷ BREAKING ON UPGRADE

   • !Consumers building from ghcr.io/purpleailab/decepticon-litellm:1.1.29 must add cache_read_input_token_cost and cache_creation_input_token_cost to any auth/* model_info overriding the built-in cost map, or cache tokens will price at $0.
8. advisory CVE-2026-61732 Jun 24, 2026 · issue 128
   • CVE-2026-61732: Role-boundary forgery via ChatML special-token literals in web crawl output composed into LLMLLMA 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. context (CVSSCVSSCommon Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. 10.0, affects <= 1.1.16)⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
9. v1.1.16 Jun 21, 2026 · release notes

   Decepticon v1.1.16 adds per-engagement sandbox routing, opt-in telemetry, web research tools, and skill retrieval.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v1.1.16 https://github.com/PurpleAILAB/Decepticon.git
   # already have the repo? check out this version:
   $ git checkout v1.1.16

   • get_sandbox(config) resolves configurable.sandbox_url and sandbox_token to enable multi-tenant sandbox isolation for bash sessions
   • Opt-in usage-telemetry system with consent gateway capturing engagement signals and masked reasoning trajectories
   • New RoE-gated open-web research tools: web_search, detect_tech_stack, live JS secret scanner, osint_enrich, and network reconnaissance suite
   • find_skill command enables LightRAG-style hybrid skill retrieval for mobile, IoTIoTA collective label for internet-connected physical devices — sensors, cameras, routers, appliances — that tools target because they run embedded firmware with limited patching and broad network exposure., and wireless security domains
   • export-transcript command exports engagement transcripts with zip export/import support for workspaces

   +1 more show less

   • New MCPMCPModel 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. server dedicated to engagement management

   6 details in Everything

10. v1.1.15 Jun 14, 2026 · release notes

    Decepticon v1.1.15 raises LLMLLMA 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. output cap to 16384 tokens and adds full LiteLLM provider catalog.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.1.15 https://github.com/PurpleAILAB/Decepticon.git
    # already have the repo? check out this version:
    $ git checkout v1.1.15

    • The factory now sets an explicit max_tokens default of 16384 (env-overridable via DECEPTICON_LLM_MAX_TOKENS)
    • litellm_dynamic_config.py now carries the full LiteLLM v1.89.0 provider catalog

    2 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !#677 drops the now-dead bedrock/vertex_ai bearer-key map entries
11. v1.1.7 Jun 6, 2026 · release notes

    Decepticon v1.1.7 adds BloodHound CE REST support, AI attack-surface detection, and sensitive TLDTLDThe rightmost label in a domain name (e.g., .com, .gov, .uk), maintained by IANA's root zone. Cyber tools filter or pivot on TLDs to scope scans, flag suspicious registrations, or classify traffic by geography or type. opt-in

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.1.7 https://github.com/PurpleAILAB/Decepticon.git
    # already have the repo? check out this version:
    $ git checkout v1.1.7

    • Adds bhce_status, bhce_cypher, and bhce_ingest_zip LangChain tools for official BHCE v9.2.2 integration
    • Introduces Technology node kind in the knowledge graphknowledge graphA representation of entities and the relationships between them. Software can follow those connections to answer questions or combine information from different sources. for AI-runtime and web-tech detection
    • Enables Blue Cell detection coverage via DetectionFired events through BlueCellTap + RuleMatcher
    • Requires explicit allow_sensitive_tlds engagement configuration for engagements with sensitive TLDs
    • Adds DECEPTICON_LLM__TIMEOUT environment variable for whole-coroutine LLMLLMA 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. timeout control

    +2 more show less

    • Deprecates legacy AD_TOOLS surface including bh_ingest_zip, bh_ingest_json, and others in favor of bhce_* tools
    • Changes BloodHound NodeKind to ADUser, ADComputer, ADGroup, etc., prefixed with AD

    7 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The bh_ingest_zip, bh_ingest_json, dcsync_check, delegation_audit, gpo_audit, adcs_audit, and adcs_post.py paths emit a DeprecationWarning; the legacy AD_TOOLS surface is deprecated.
    • !Engagements with sensitive TLDs in scope now require explicit opt-in via allow_sensitive_tlds; previously trusted engagements will fail scope evaluation without this flag.
12. v1.1.0 May 21, 2026 · release notes

    Decepticon v1.1.0 adds plugin override system, runtime /plugins and /agent slash commands, HTTP-only sandbox, and Soundwave 8-doc bundle expansion

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.1.0 https://github.com/PurpleAILAB/Decepticon.git
    # already have the repo? check out this version:
    $ git checkout v1.1.0

    • decepticon.bundles entry-point group for declaring custom PluginBundle packages that override agent factories, roles, middleware, and prompts without forking the core code.
    • MiddlewareSlot.ENGAGEMENT_CONTEXT, MiddlewareSlot.SKILLS constants and build_middleware() function enable plugins to inject slots and skill sources directly in Python code.
    • All 16 agent factories now accept tools=, middleware=, and system_prompt= kwargs with full-replace semantics for composing custom agent pipelines.
    • LLMFactory.get_model(role, default_role=...) lets plugin orchestrators inherit an OSSOSSOpen-source software: code whose licence permits anyone to inspect, modify, and redistribute it. Cyber tools build on it to reuse auditable libraries without licensing fees. role's model assignment until they ship their own.
    • The HTTPSandbox class replaces the removed DockerSandbox; all sandbox interaction now flows through HTTPSHTTPSHTTP carried over an encrypted TLS connection. It protects traffic in transit and lets a browser verify the server's certificate. over the management network, removing the need for a docker.sock mount.

    +3 more show less

    • Two new slash commands: /plugins to list/enable/disable plugin bundles at runtime without restart, and /agent to switch the active orchestrator between OSS standard, vulnresearch, or any plugin-registered agent.
    • Long-running shell tools (e.g. nmap scans) now emit ● Background command finished — <output> events automatically upon completion.
    • Soundwave engagement planning interview now produces eight structured documents: plan/roe.json, plan/threat-profile.json, plan/conops.json, plan/deconfliction.json, plan/contact.json, plan/data-handling.json, plan/abort.json, and plan/cleanup.json.

    8 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The internal decepticon.agents.assembly module has been renamed to decepticon.agents.build.
13. v1.0.16 May 3, 2026 · release notes

    Decepticon v1.0.16 adds Ollama support, DeepSeek V4 models, new hunting lanes, and make dogfood workflow

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.16 https://github.com/PurpleAILAB/Decepticon.git
    # already have the repo? check out this version:
    $ git checkout v1.0.16

    └──▷ TRY IT
    Build a Docker image with a specific version tag for release automation.
    copy

    $ docker build --build-arg VERSION=v1.0.16 .

    Run the OSSOSSOpen-source software: code whose licence permits anyone to inspect, modify, and redistribute it. Cyber tools build on it to reuse auditable libraries without licensing fees. UX validation flow in an isolated environment to verify launcher and CLI behavior.
    copy

    $ make dogfood

    • Adds --build-arg VERSION=<tag> to Docker builds for release flow automation and make dogfood command for OSS UX validation in .dogfood/ directory
    • Introduces DeepSeek V4 models with v4-pro and v4-flash profiles supporting 1M context, reasoning, and tool calls
    • Adds three new analyst hunting lanes: Trust-boundary (Lane F), Pattern-exhaustion (Lane G), and Bounty target assessment (Lane H)
    • Integrates two new CLI tools: bounty_scope_check and format_bounty_report
    • Deploys new skill file skills/exploit/web/smuggling.md for Class A vs Class B HRS branching analysis

    +1 more show less

    • Updates launcher client (clients/launcher/cmd/ollama_models.go) to use parallel fan-out for /api/tags and /api/show endpoints

    6 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Removes make sync-version command from the build flow
14. v1.0.12 Apr 30, 2026 · release notes

    Adds OpenRouterOpenRouterA 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. and NVIDIA NIM LLMLLMA 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. support, parallel task dispatch, and macOS Apple Silicon compatibility.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.12 https://github.com/PurpleAILAB/Decepticon.git
    # already have the repo? check out this version:
    $ git checkout v1.0.12

    • Adds OPENROUTER_API_KEY and NVIDIA_API_KEY environment variables in the onboarding wizard to enable Claude via OpenRouter and Llama/Nemotron via NVIDIA NIM
    • Enables parallel execution of independent objectives by defaulting the orchestrator to concurrent task() calls for goals without blocked_by dependencies
    • Orders credential fallback chains based on configured availability, controlled by the DECEPTICON_AUTH_PRIORITY environment variable
    • Adds macOS Apple Silicon compatibility by forcing linux/amd64 platform in docker-compose.yml for Kali-based services and target containers

    4 details in Everything

15. v1.0.9 Apr 26, 2026 · release notes

    Adds XBOW benchmark harness, compose --wait, env-driven version pinning, and new env vars.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.9 https://github.com/PurpleAILAB/Decepticon.git
    # already have the repo? check out this version:
    $ git checkout v1.0.9

    └──▷ TRY IT
    Extend startup timeout for slow hardware before running the cyber tool.
    copy

    $ DECEPTICON_STARTUP_TIMEOUT_SECONDS=120 decepticon

    Ensure clean disk cleanup when uninstalling the cyber tool.
    copy

    $ decepticon remove --volumes --remove-orphans

    • DECEPTICON_STARTUP_TIMEOUT_SECONDS environment variable sets the agent stack startup timeout (default 600s).
    • --volumes --remove-orphans flags added to decepticon remove to free disk space on uninstall.
    • BENCHMARK_MODE environment variable injected by the new XBOW benchmark framework.
    • New decepticon.benchmark package with provider, harness, and scorer components for end-to-end evaluation.

    4 details in Everything

16. v1.0.8 Apr 23, 2026 · release notes

    Adds StateGraph-based engagement lifecycle router with conops → opplan → execution → report phase transitions

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.8 https://github.com/PurpleAILAB/Decepticon.git
    # already have the repo? check out this version:
    $ git checkout v1.0.8

17. v1.0.7 Apr 22, 2026 · release notes

    Decepticon v1.0.7 introduces Go CLI binary, MiniMax provider, Claude OAuth, interactive wizard, and dashboard sync

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.7 https://github.com/PurpleAILAB/Decepticon.git
    # already have the repo? check out this version:
    $ git checkout v1.0.7

    └──▷ USE IT
    Run docker compose in CI/CD to skip local image builds using the new flag
    shell copy

    docker compose run --no-build decepticon

    • decepticon is now a native Go binary with an Interactive onboard wizard (Charmbracelet Huh v2)
    • Native cross-platform binaries for Linux (amd64/arm64), macOS (amd64/arm64), and Windows
    • Bash fallback launcher at scripts/launcher.sh remains available if Go binary is missing
    • Install scripts (install.sh, Go launcher, bash launcher) pin downloads to the specific release v{version} tag
    • Added DECEPTICON_MODEL_PROVIDER=auth environment variable for Claude Code OAuth Gateway access

    +4 more show less

    • ModelProvider and ModelProfile configuration fields are now separated
    • MiniMax added as a fourth LLMLLMA 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 alongside Anthropic, OpenAI, and Google
    • Web Dashboard gains Octogent integration for workspace sync, plan viewer, and agent canvas
    • Dashboard supports thread persistence to resume sessions across restarts

    9 details in Everything

18. v1.0.6 Apr 18, 2026 · release notes

    Adds Human-in-the-Loop pause/resume via Ctrl+C and /resume command with message queuing

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.6 https://github.com/PurpleAILAB/Decepticon.git
    # already have the repo? check out this version:
    $ git checkout v1.0.6

    • Enables pausing an agent run at a checkpoint with Ctrl+C, preserving all state
    • Adds /resume [feedback] subcommand to continue a paused agent run with optional operator feedback
    • Introduces message queuing during agent streaming — typed messages are auto-submitted on completion

    3 details in Everything

19. v1.0.4 Apr 18, 2026 · release notes

    Decepticon v1.0.4 adds web dashboard, offensive vaccine agents, and OPPLAN persistence.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.4 https://github.com/PurpleAILAB/Decepticon.git
    # already have the repo? check out this version:
    $ git checkout v1.0.4

    └──▷ TRY IT
    Run Decepticon with a custom port for the new default web dashboard.
    copy

    $ WEB_PORT=8080 decepticon

    Disable the Claude 4.x prompt compatibility shim if needed.
    copy

    $ DECEPTICON_CLAUDE4_COMPAT=0 decepticon

    Load an existing operational plan from a specific workspace directory.
    code copy

    load_opplan(workspace_path)

    • Web dashboard is now in the default stack with WEB_PORT env var override
    • Adds 12 new specialized agents for the Offensive Vaccine loop including defender, detector, and exploiter
    • OPPLAN persistence via plan/opplan.json with save_opplan / load_opplan tools
    • Claude 4.x compatibility layer via DECEPTICON_CLAUDE4_COMPAT=0 opt-out env var
    • SARIFSARIFStatic Analysis Results Interchange Format, a standard for sharing code-analysis findings, including locations and rule details. It lets scanners report results to a common review interface. vulnerability findings ingest into the Knowledge Graph

    5 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Python direct execution entry point has been removed in favor of the web dashboard
20. v1.0.3 Apr 6, 2026 · release notes

    Decepticon v1.0.3 introduces Makefile shortcuts, structured finding models, and editable Docker development workflow.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.3 https://github.com/PurpleAILAB/Decepticon.git
    # already have the repo? check out this version:
    $ git checkout v1.0.3

    • Adds make dev, make cli, make start, make test, and make lint targets to the new Makefile for streamlined local development
    • Introduces decepticon command entry point that launches a full Docker stack matching production environments
    • Adds Pydantic models for Finding, Evidence, and AttackPath with CVSSCVSSCommon Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. v4.0 scoring and confidence levels
    • Implements new engagement directory layout with findings/, findings/evidence/, findings/attack-paths/, report/, and timeline.jsonl
    • Adds allowed-tools field to skill frontmatter for progressive tool disclosure

    +1 more show less

    • Enables Docker Compose Watch for automatic source-to-container syncing without manual rebuilds

    6 details in Everything

21. v1.0.2 Apr 4, 2026 · release notes

    Adds Soundwave agent for engagement docs, OPPLANMiddleware CRUDCRUDThe 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. tools, dual-screen CLI with ctrl+o toggle

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.2 https://github.com/PurpleAILAB/Decepticon.git
    # already have the repo? check out this version:
    $ git checkout v1.0.2

    • Dual-screen CLI adds ctrl+o toggle between Prompt mode (compact) and Transcript mode (full history)
    • Adds OPPLANMiddleware with domain-specific CRUD tools for objectives (add/get/list/update) including state transition validation and dependency checking
    • Introduces DecepticonSkillsMiddleware, a MITRE ATT&CKMITRE ATT&CKA publicly maintained knowledge base of adversary tactics and techniques observed in real attacks, maintained by MITRE. Cyber tools use it as a common taxonomy to classify, map, and communicate threat behavior.-aware skill system supporting phase grouping and progressive disclosure
    • Soundwave agent replaces the deprecated Planner agent for engagement document generation (RoE, ConOps, Deconfliction Plan, OPPLAN)
    • SafeCommandMiddleware expands its blocklist to include session-destroying commands such as pkill, nsenter, and docker exec

    +1 more show less

    • PromptBuilder introduces static/dynamic prompt separation with cache boundary markers and cross-cutting patterns

    6 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The skills/planning/ directory is renamed to skills/soundwave/
    • !Model profiles DEFAULT and HIGH are renamed to eco and max respectively
22. v1.0.1 Mar 30, 2026 · release notes

    Decepticon v1.0.1 adds modular C2 infrastructure via Docker profiles and async sandbox execution.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.1 https://github.com/PurpleAILAB/Decepticon.git
    # already have the repo? check out this version:
    $ git checkout v1.0.1

    • Activates Sliver C2 in c2-sliver container using docker compose --profile c2-sliver
    • Extracts shared prompt fragments (bash.md, skills.md) for reuse via load_prompt()
    • Adds execute_tmux_async() method to DockerSandbox for non-blocking tmux execution

    3 details in Everything

23. v1.0.0 Mar 25, 2026 · release notes

    Decepticon v1.0.0 releases as an autonomous red team framework with multi-agent orchestration and streaming CLI

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v1.0.0 https://github.com/PurpleAILAB/Decepticon.git
    # already have the repo? check out this version:
    $ git checkout v1.0.0

    └──▷ USE IT
    Install and launch the autonomous red team framework from the command line
    shell copy

    curl -fsSL https://raw.githubusercontent.com/PurpleAILAB/Decepticon/main/scripts/install.sh | bash && decepticon

    • Adds decepticon CLI entrypoint to launch the autonomous red team loop named Ralph
    • Introduces orchestrator that delegates to specialized sub-agents for recon, planner, exploit, and post-exploit phases
    • Supports Kali Linux sandbox via isolated Docker container with pre-installed security tools (nmap, hydra, sqlmap, nikto, gobuster)
    • Adds Streaming CLI named Ink for real-time visibility into sub-agent tool calls and bash execution
    • Enables dual LLMLLMA 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. modes: API key via LiteLLM proxy or OAuth for Claude Pro/Max, ChatGPT Plus, and GitHub Copilot

    5 details in Everything