No entries match these filters. Clear filters
OpenHands v1.24.0 adds workspace folder toggling, OpenAI subscription cache scoping, and read-only shared automation conversations
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.24.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout v1.24.0
└──▷ HOW TO FIND IT
Quickly enable or disable all workspace folders at once via the Conversations header.
📍 Click the toggle icon in the Conversations header to select/deselect all workspace folders.
Ensure OpenAI subscription caches are isolated per backend to prevent cross-tenant data leakage.
📍 N/A - This is a backend infrastructure change applied automatically.
View shared automation conversations on cloud without risk of accidental modifications.
📍 Open a shared automation conversation link; the UI will enforce read-only mode.
Adds ability to toggle all workspace folders from the Conversations header Not checked yet Scopes OpenAI subscription caches by backend instead of globally Not checked yet Opens shared automation conversations as read-only on cloud deployments Not checked yet
3 details in Everything
Adds universal macOS DMG installer, light themes, and deployment-kind telemetry tags.
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.23.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout v1.23.0
└──▷ HOW TO FIND IT
Install on macOS with native ARM ARM A family of RISC processor architectures designed by Arm Holdings and licensed to chip makers worldwide; tools target it to run natively on mobile, embedded, and increasingly server and desktop hardware. and Intel binaries in a single package.
📍 Download the universal macOS DMG from the release assets and mount it.
Switch to a light theme for better visibility in bright environments.
📍 In the settings or appearance menu, select Light+ or Solarized Light .
universal macOS DMG with per-arch bundled runtimes Not checked yet Light+ and Solarized Light theme support in the UI Not checked yet Agent Canvas telemetry tagged by deployment kind Not checked yet
3 details in Everything
v1.22.0 adds remote 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. server testing on cloud backends and Git Sync access for org admins
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.22.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout v1.22.0
Add ability to test remote MCP servers on cloud backends via the app server Not checked yet Open Git Sync configuration to organization administrators on cloud backends Not checked yet
2 details in Everything
Adds ability to request Docker execution workspaces from remote Docker servers.
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.21.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout v1.21.0
OpenHands v1.20.0 adds agent profiles to select secrets and saved automations
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.20.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout v1.20.0
agent-profiles feature enables selecting specific secrets available to a profile Not checked yet
1 detail in Everything
OpenHands v1.19.0 adds GPT-6 Astra model support, 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. server scoping to agent profiles, and automation disablement visibility.
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.19.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout v1.19.0
Adds GPT-6-Astra to the list of supported 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 for agent execution Not checked yet Enables scoping an agent profile to specific MCP servers via a new configuration option in the agent settings UI Not checked yet Surfaces the automation disablement reason in the Agent Canvas panel when automations are turned off Not checked yet
3 details in Everything
OpenHands v1.18.0 adds automation visibility, cloud editing, and Canvas model flags
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.18.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout v1.18.0
└──▷ HOW TO FIND IT
Verify which identity is executing your automations to ensure correct permissions.
📍 In the OpenHands web UI, check the automation details panel for the runs as identity field.
Modify an existing automation when operating on a cloud backend host.
📍 Navigate to the Automations list in the OpenHands UI and click Edit on a cloud-hosted automation entry.
Configure default model usage via DB-driven flags in the Canvas interface.
📍 Access Canvas settings and update the DB-driven free/default model flags to control inference costs.
Adds show which identity an automation runs as to expose the execution identity for automations. Not checked yet Enables editing automations directly on cloud backends via allow editing automations on cloud backends. Not checked yet Restricts automation re-enabling so only the creator can turn it back on with only the creator can turn an automation back on. Not checked yet Adds DB-driven free/default model flags in Canvas controlled by the Canvas subsystem. Not checked yet
4 details in Everything
OpenHands v1.17.0 adds conversation tags and filtering, local Agent Canvas Planner support, and generic custom automation manifests
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.17.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout v1.17.0
└──▷ HOW TO FIND IT
Filter your conversation list by tags to quickly find related automations or discussions.
📍 In the conversation panel, use the new tag filter dropdown to select and narrow results by automation or custom tags.
Configure a recurring automation with a precise schedule using a standard cron string directly in the UI.
📍 Open the edit-automation modal and enter your schedule (e.g., '0 9 * * 1') into the new custom cron expression field.
Enable 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 integrations for your cloud-hosted canvas instance from the centralized settings area.
📍 Navigate to the unified cloud settings entrypoint for Agent Canvas in the application menu, then configure your provider credentials under the new integration section.
conversation-panel: add conversation tags and filtering Not checked yet enable LLM provider connections on cloud Not checked yet allow editing a custom cron expression from the edit-automation modal Not checked yet unified cloud settings entrypoint for Agent Canvas Not checked yet display message timestamps on hover Not checked yet
+3 more
show less
split automation permissions into view/manage with creator escape hatch Not checked yet show manifest-declared value statements on dashboard cards and rows Not checked yet Make automation run UI task-outcome aware Not checked yet
8 details in Everything
OpenHands v1.16.0 adds 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 selection, allow-list skills, and live automation phases
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.16.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout v1.16.0
feat(settings) : select supported LLM providers Not checked yet feat(automations) : show a run's live phase Not checked yet feat(settings) : expose the LLM-switching toggle in Agent settings Not checked yet feat(skills) : replace the all-on skill catalog with an explicit allow-list Not checked yet
4 details in Everything
OpenHands v1.15 adds 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 connections UI, automated script bundle installs via automation catalog, and a workspace path view in the Files pane
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.15.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout v1.15.0
└──▷ HOW TO FIND IT
Configure a local agent-server model connection via the new UI panel without editing environment variables.
📍 Navigate to Settings › LLM provider-connections and select local agent-server from the dropdown.
Deploy a new automation that includes scripts by pulling a catalog entry instead of manually uploading files.
📍 In the Automations dashboard, select Add and choose a catalog entry that includes a script bundle.
Adds an LLM provider-connections UI (local agent-server) to configure model endpoints directly from the client interface. Not checked yet Enables automations to install a catalog entry that ships a script bundle automatically during deployment. Not checked yet Shows the workspace path in the Files view for easy navigation and copying of the active directory location. Not checked yet
3 details in Everything
OpenHands Activity Log now shows 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. cost in USD and lets you export run data as CSV or JSON.
Activity Log displays the LLM cost (in USD, four decimal places) for each completed run; a zero cost shows as $0.0000 and an unavailable cost shows nothing. Not checked yet New Activity Log export controls let you download run data as CSV or JSON, each including a raw numeric cost field per run (exported as null when the backend reports no cost). Not checked yet Automate pane in the sidebar lists all automations on the active backend with name, trigger type, and enabled state, and surfaces an add option when no automations exist yet. Not checked yet
3 details in Everything
OpenHands v1.14.0 adds Git Sync automation, 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. pre-flight validation, structured Canvas error display, and Kimi K3 as the free default Canvas model.
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.14.0 https://github.com/OpenHands/OpenHands.git
$ git checkout v1.14.0
Adds a Git Sync page under Automations for managing repository synchronization workflows. Not checked yet Adds LLM pre-flight validation that checks and rejects misconfigured LLM profiles before they can be saved. Not checked yet Canvas now presents structured error outcomes so failures surface with actionable detail rather than raw errors. Not checked yet Sets Kimi K3 as the default Canvas model and tags it as free, giving users a no-cost starting point. Not checked yet
4 details in Everything
OpenHands SDK launches with modular AI agent framework supporting bash, file editing, web browsing, and LiteLLM-compatible 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. backends
└──▷ USE IT
Authenticate with a ChatGPT Plus/Pro subscription to run an agent on Codex without using API credits.
python
copy
from openhands.sdk import LLM
llm = LLM.subscription_login(vendor="openai", model="gpt-5.2-codex")
Create a minimal agent that writes project facts to a file, using OpenHands Cloud as the LLM backend.
copy
$ export LLM_MODEL="openhands/claude-sonnet-4-5-20250929"
export LLM_API_KEY="your-openhands-api-key"
uv run python examples/01_standalone_sdk/01_hello_world.py
Compose and run an agent programmatically with terminal and file-editing capabilities in the current working directory.
python
copy
import os
from openhands.sdk import LLM, Agent, Conversation, Tool
from openhands.tools.file_editor import FileEditorTool
from openhands.tools.terminal import TerminalTool
llm = LLM(
model=os.getenv("LLM_MODEL", "anthropic/claude-sonnet-4-5-20250929"),
api_key=os.getenv("LLM_API_KEY"),
base_url=os.getenv("LLM_BASE_URL", None),
)
agent = Agent(llm=llm, tools=[Tool(name=TerminalTool.name), Tool(name=FileEditorTool.name)])
conversation = Conversation(agent=agent, workspace=os.getcwd())
conversation.send_message("List all Python files and count lines of code.")
conversation.run()
Introduces openhands-sdk and openhands-tools PyPI packages (must be installed together at matching versions, e.g. pip install "openhands-sdk==1.22.1" "openhands-tools==1.22.1") for building AI agents that execute bash commands, edit files, and browse the web. Not checked yet Optional openhands-workspace and openhands-agent-server packages add sandboxed workspaces in Docker or remote servers. Not checked yet Adds LLM class accepting model, api_key, and base_url constructor arguments, supporting any LiteLLM-compatible provider via the LLM_MODEL, LLM_API_KEY, and LLM_BASE_URL environment variables. Not checked yet Adds LLM.subscription_login(vendor, model) method to authenticate with a ChatGPT Plus/Pro subscription and access Codex models (e.g. gpt-5.2-codex) without consuming API credits. Not checked yet Supports openhands/-prefixed model names (e.g. openhands/claude-sonnet-4-5-20250929) when using OpenHands Cloud as the LLM provider. Not checked yet
+5 more
show less
Provides built-in tools — TerminalTool, FileEditorTool, and TaskTrackerTool — importable from openhands.tools.terminal, openhands.tools.file_editor, and openhands.tools.task_tracker. Not checked yet Adds Agent class accepting llm and tools arguments to compose an agent from an LLM backend and a list of Tool objects. Not checked yet Adds Conversation class with send_message() and run() methods to manage the interaction lifecycle between caller and agent within a given workspace path. Not checked yet Ships 24+ runnable examples under examples/01_standalone_sdk/, covering custom tools (02_custom_tools.py), microagent skills (03_activate_microagent.py), and more. Not checked yet Supports installation from source via git clone https://github.com/OpenHands/software-agent-sdk.git and make build. Not checked yet
10 details in Everything
OpenHands adds Agent Canvas with ACP support, Sandbox Server REST API REST API A web API design style where clients interact with resources over HTTP using standard methods like GET, POST, and DELETE, giving cyber tools a predictable, language-agnostic way to integrate with external services. v1, 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. integrations, and org-level spend budgets.
New Agent Canvas component connects to execution, automation, and sandbox services, with documented architecture for understanding how the pieces fit together. Not checked yet Supports running Claude Code, Codex, or Gemini CLI inside Agent Canvas via the Agent Client Protocol (ACP). Not checked yet New Sandbox Server REST API (V1) exposes REST endpoints for conversations and sandboxes. Not checked yet Agent Canvas can be connected to a remote Agent Server backend running on another machine or container. Not checked yet Agent Canvas can be run with Docker for a sandboxed backend with a mounted project workspace. Not checked yet
+4 more
show less
Adds LLM profile management in Agent Canvas — configure models and reuse saved LLM profiles across conversations. Not checked yet Adds support for connecting OpenHands to local LLM servers including LM Studio, Ollama, vLLM, and SGLang. Not checked yet Adds org-level spend budgets to cap AI spending for organizations and their members. Not checked yet New Skills system lets practitioners give OpenHands reusable instructions, domain knowledge, workflows, and supporting resources. Not checked yet
9 details in Everything
OpenHands v1.13.0 adds conversation archiving, inline markdown artifact previews, a 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. usage meter with manual compaction, and an issue readiness gate.
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.13.0 https://github.com/OpenHands/OpenHands.git
$ git checkout v1.13.0
Adds client-side conversation archive, letting users archive and manage past conversations directly in the UI. Not checked yet Adds inline markdown artifact previews in the chat panel, rendering markdown output without leaving the conversation. Not checked yet Adds a context window usage meter, usage drawer, and manual compaction control so users can monitor and manage 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. context consumption mid-session. Not checked yet Adds a ready-for-dev issue readiness gate with type-specific criteria to validate issues before they are handed off to the agent. Not checked yet
4 details in Everything
OpenHands v1.11.0 adds per-run 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. cost visibility, tag filtering for automations, and typed child-conversation actions.
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.11.0 https://github.com/OpenHands/OpenHands.git
$ git checkout v1.11.0
Shows per-run LLM cost in the Activity Log and exports, giving teams immediate spend visibility for each agent run. Not checked yet Adds automation tag filter and recognition so automations can be organized and surfaced by tag. Not checked yet Adds tag chips, overflow handling, and hovercard labels to the conversations list for richer conversation metadata display. Not checked yet Adds a typed agent action for launching local or Cloud child conversations, enabling programmatic multi-conversation workflows. Not checked yet Keeps the Automations pane visible even when empty, reducing navigation confusion. Not checked yet
+2 more
show less
Reorders the Customize navigation section for improved discoverability. Not checked yet Polishes the Agent Canvas version-update UI in Settings. Not checked yet
7 details in Everything
OpenHands v1.10.0 adds activity log export, a featured automations dashboard, manifest-driven automation sub-pages, and faceted skills filtering.
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.10.0 https://github.com/OpenHands/OpenHands.git
$ git checkout v1.10.0
Adds activity log export capability to the interface. Not checked yet Ships a featured automations landing dashboard for discovering and launching automations. Not checked yet Adds manifest-driven sub-pages to the automation interface, enabling structured automation workflows. Not checked yet Adds faceted rail filtering to the Skills page for faster skill discovery. Not checked yet Sets GLM 5.2 as the default model for Canvas. Not checked yet
5 details in Everything
OpenHands agent canvas now lets you inspect plugin contents including bundled skills and a syntax-highlighted file viewer.
Adds a plugin detail view in the agent canvas that lists bundled skills (including command-derived skills) with their icons, names, and descriptions. Not checked yet Adds an expandable directory tree in the plugin detail view with inline syntax-highlighted file viewing — select a file to open it, select again to close. Not checked yet
2 details in Everything
OpenHands 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. profiles can now be assigned specifically for conversation title generation
Adds a dedicated LLM profile selector for conversation title generation under Settings > Application > Conversation titles, with an Automatic option to use the active local profile or a specific saved profile. Not checked yet
1 detail in Everything
OpenHands Agent Canvas ships a preview desktop app for macOS (Apple silicon) and Windows that bundles all runtimes.
New Agent Canvas desktop app Preview for macOS (Apple silicon) and Windows bundles Node.js and required runtimes — no prerequisites or open terminal needed; download Agent-Canvas-<version>-arm64.dmg (macOS) or Agent-Canvas-Setup-<version>.exe (Windows) from the OpenHands releases page. Not checked yet Desktop app auto-starts its local backend on launch; select Show details during startup to view and copy the live startup log for troubleshooting. Not checked yet Supports in-app update checking and version display via Settings > Application. Not checked yet
3 details in Everything
OpenHands offers a native desktop app preview build for testing.
OpenHands v1.9.0 adds live agent activity in chat, a compact backend chooser, and an extension-manifest hosting system.
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.9.0 https://github.com/OpenHands/OpenHands.git
$ git checkout v1.9.0
Shows live agent activity directly in the chat view so users can monitor what the agent is doing in real time. Not checked yet Adds a compact Cloud vs Agent-server backend chooser in the UI for faster backend selection. Not checked yet Introduces a domain-neutral extension-manifest host, enabling extensions to be served without domain-specific coupling. Not checked yet
3 details in Everything
OpenHands v1.8.0 lets you toggle installed 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. servers on or off directly from their cards.
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.8.0 https://github.com/OpenHands/OpenHands.git
$ git checkout v1.8.0
Adds enable/disable toggle for installed MCP servers directly from the server card in the UI. Not checked yet
1 detail in Everything
OpenHands v1.7.0 adds persistent agent memory, always-visible 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. selector, and secret overwriting from the edit form.
└──▷ GET THIS VERSION
copy
$ git clone --branch v1.7.0 https://github.com/OpenHands/OpenHands.git
$ git checkout v1.7.0
Adds a persistent agent memory toggle so agents can retain context across sessions. Not checked yet LLM selector is now always visible in the UI; agent-profile switching moves to the tools menu. Not checked yet Enables overwriting an existing secret's value directly from the secret edit form. Not checked yet
3 details in Everything
OpenHands Agent Canvas adds saved 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. profiles and an in-chat /model command for switching models mid-conversation.
└──▷ TRY IT
Switch to a faster or cheaper model mid-task without losing conversation context or workspace state.
copy
$ /model fast-profile
Connect Agent Canvas to a local vLLM or LM Studio server running inside Docker by entering the correct backend-reachable URL in the Advanced tab.
yaml
copy
Model: openai/<served-model-id>
Base URL: http://host.docker.internal:1234/v1
API key: local-llm
Route requests through a LiteLLM proxy by using the required model name prefix and proxy credentials in the Advanced tab.
yaml
copy
Model: litellm_proxy/<model-name>
Base URL: <your-litellm-proxy-base-url>
API key: <your-proxy-api-key>
Adds saved LLM profiles in Settings > LLM, letting practitioners store named provider/model/credential sets and switch between them without re-entering API keys. Not checked yet Supports Basic, Advanced, and All configuration tabs for direct providers, OpenHands LLM API keys, local OpenAI-compatible servers (LM Studio, Ollama, vLLM, etc.), and LiteLLM proxies. Not checked yet Introduces /model chat command to list available profiles (/model) or switch to a named profile (/model <profile-name>) mid-conversation while preserving history, workspace, and task state. Not checked yet Adds model-switch events to the conversation timeline so practitioners can audit when and which profile was applied during a task. Not checked yet Links Agent Profiles to LLM profiles, allowing each agent profile to reference a specific model configuration via Settings > Agent. Not checked yet
5 details in Everything
OpenHands Agent Canvas adds a prebuilt GitHub PR Review workflow that watches pull requests and posts automated reviews.
└──▷ HOW TO FIND IT
Set up the prebuilt PR review automation to watch a GitHub repository and post agent-generated reviews on new pull requests.
📍 In Agent Canvas: open Automate › Start from a proven workflow › choose the GitHub PR review workflow. Send the prefilled prompt (or edit it to specify the repository owner/name, which PRs to review, and whether to post inline or summary feedback).
Connect the GitHub MCP Server MCP Server A process that exposes tools, resources, and prompts over Anthropic's Model Context Protocol, letting an AI agent call external APIs or local services through a standardised interface. so Agent Canvas agents can read PRs and post review comments using a fine-grained personal access token personal access token A personal access token, a credential used by scripts and integrations instead of a user's password. Its permissions and expiry depend on the service issuing it. .
📍 In Agent Canvas: open Customize › MCP Servers › select GitHub from the MCP library › paste your fine-grained token (with Contents/Issues/Pull requests read+write and Metadata read-only) › enable the secret-creation toggle › Save.
New prebuilt GitHub PR Review workflow in Agent Canvas lets an OpenHands agent automatically review pull requests and post summary or inline feedback. Not checked yet GitHub 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. Server integration in Agent Canvas provides the agent tools for reading repositories, inspecting PRs, and posting review output. Not checked yet Configurable review scope lets the automation target specific repositories, post single summary comments or detailed inline feedback, inspect CI/check results, and ignore specified files or directories. Not checked yet
3 details in Everything
OpenHands Agent Canvas adds a prebuilt GitHub PR Review workflow that watches pull requests and posts automated reviews.
└──▷ HOW TO FIND IT
Set up the prebuilt PR review automation to watch a GitHub repository and post agent-generated reviews on new pull requests.
📍 In Agent Canvas: open Automate › Start from a proven workflow › choose the GitHub PR review workflow. Send the prefilled prompt (or edit it to specify the repository owner/name, which PRs to review, and whether to post inline or summary feedback).
Connect the GitHub MCP Server MCP Server A process that exposes tools, resources, and prompts over Anthropic's Model Context Protocol, letting an AI agent call external APIs or local services through a standardised interface. so Agent Canvas agents can read PRs and post review comments using a fine-grained personal access token personal access token A personal access token, a credential used by scripts and integrations instead of a user's password. Its permissions and expiry depend on the service issuing it. .
📍 In Agent Canvas: open Customize › MCP Servers › select GitHub from the MCP library › paste your fine-grained token (with Contents/Issues/Pull requests read+write and Metadata read-only) › enable the secret-creation toggle › Save.
New prebuilt GitHub PR Review workflow in Agent Canvas lets an OpenHands agent automatically review pull requests and post summary or inline feedback. Not checked yet GitHub 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. Server integration in Agent Canvas provides the agent tools for reading repositories, inspecting PRs, and posting review output. Not checked yet Configurable review scope lets the automation target specific repositories, post single summary comments or detailed inline feedback, inspect CI/check results, and ignore specified files or directories. Not checked yet
3 details in Everything
OpenHands adds Agent Canvas UI, persistent memory, path-triggered rules, SDK plugins, and full Kubernetes/Enterprise deployment docs.
└──▷ HOW TO FIND IT
Install Agent Canvas into an existing Kubernetes cluster for a team deployment.
📍 Follow the Helm installation guide at docs.openhands.dev/openhands/usage/agent-canvas/backend-setup/kubernetes.md to add the official Helm chart and configure your values.yaml.
Enable persistent memory for an agent so it retains context across separate conversations.
📍 In your SDK agent configuration, enable the two-tier persistent memory option as described at docs.openhands.dev/sdk/guides/persistent-memory.md.
New Agent Canvas: a desktop UI for running agents with reusable profiles, conversation branching, and an Automate view for managing automations. Not checked yet Adds opt-in two-tier persistent memory that survives across conversations via the SDK. Not checked yet Introduces path-triggered rules that deterministically inject context whenever an agent reads, edits, or creates a file matching a glob pattern. Not checked yet Supports plugin discovery, installation, and attachment directly inside Agent Canvas. Not checked yet Enables custom sandbox images so agents start with pre-loaded repos, dependencies, and tooling. Not checked yet
+5 more
show less
Adds Docker-in-sandbox capability, letting agents run containers, Docker Compose, and image builds without privileged cluster access. Not checked yet New Azure DevOps integration for authentication and automation triggers in OpenHands Enterprise. Not checked yet Adds Laminar-based analytics for trace analysis in OpenHands Enterprise. Not checked yet Supports mobile access to Agent Canvas via Tailscale or ngrok. Not checked yet Enterprise Skills and Plugins management for repository, organization, and user-scoped skills. Not checked yet
10 details in Everything
OpenHands adds Agent Canvas UI, persistent memory, path-triggered rules, SDK plugins, and full Kubernetes/Enterprise deployment docs.
└──▷ HOW TO FIND IT
Install Agent Canvas into an existing Kubernetes cluster for a team deployment.
📍 Follow the Helm installation guide at docs.openhands.dev/openhands/usage/agent-canvas/backend-setup/kubernetes.md to add the official Helm chart and configure your values.yaml.
Enable persistent memory for an agent so it retains context across separate conversations.
📍 In your SDK agent configuration, enable the two-tier persistent memory option as described at docs.openhands.dev/sdk/guides/persistent-memory.md.
New Agent Canvas: a desktop UI for running agents with reusable profiles, conversation branching, and an Automate view for managing automations. Not checked yet Adds opt-in two-tier persistent memory that survives across conversations via the SDK. Not checked yet Introduces path-triggered rules that deterministically inject context whenever an agent reads, edits, or creates a file matching a glob pattern. Not checked yet Supports plugin discovery, installation, and attachment directly inside Agent Canvas. Not checked yet Enables custom sandbox images so agents start with pre-loaded repos, dependencies, and tooling. Not checked yet
+5 more
show less
Adds Docker-in-sandbox capability, letting agents run containers, Docker Compose, and image builds without privileged cluster access. Not checked yet New Azure DevOps integration for authentication and automation triggers in OpenHands Enterprise. Not checked yet Adds Laminar-based analytics for trace analysis in OpenHands Enterprise. Not checked yet Supports mobile access to Agent Canvas via Tailscale or ngrok. Not checked yet Enterprise Skills and Plugins management for repository, organization, and user-scoped skills. Not checked yet
10 details in Everything
OpenHands cloud-1.47.0 gates Agent Canvas behind SaaS SaaS Software delivered over the internet and billed by subscription, hosted and operated by the vendor so the buyer skips installation and infrastructure management. authentication.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.47.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout cloud-1.47.0
OpenHands cloud-1.47.0 gates Agent Canvas behind SaaS SaaS Software delivered over the internet and billed by subscription, hosted and operated by the vendor so the buyer skips installation and infrastructure management. authentication.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.47.0 https://github.com/OpenHands/OpenHands.git
$ git checkout cloud-1.47.0
OpenHands 1.46 adds configurable BYOR key alias patterns and enriched final archive manifests.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.46.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout cloud-1.46.0
Enables configurable key alias patterns for Bring Your Own Router (BYOR) keys in enterprise deployments. Not checked yet Enriches final archive manifests with additional metadata and removes initial snapshots to streamline workspace archives. Not checked yet
2 details in Everything
OpenHands 1.46 adds configurable BYOR key alias patterns and enriched final archive manifests.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.46.0 https://github.com/OpenHands/OpenHands.git
$ git checkout cloud-1.46.0
Supports configurable Bring-Your-Own-Router (BYOR) key alias patterns for enterprise deployments. Not checked yet Enriches final archive manifests with additional metadata and removes redundant initial snapshots. Not checked yet
2 details in Everything
OpenHands 1.11.0 adds Agent Profiles, a Budgets dashboard, expanded usage monitoring, and email provider visibility.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.11.0 https://github.com/OpenHands/OpenHands.git
$ git checkout 1.11.0
└──▷ HOW TO FIND IT
Verify your email provider is active before relying on agent notifications — check the surfaced 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. /Resend status in admin settings.
📍 In the console, go to Settings › Email and confirm the SMTP or Resend provider status is shown as enabled.
Monitor team spend against budget limits to catch runaway agent usage before it exceeds allocation.
📍 In the console, go to Dashboards › Budgets to view configured budget thresholds and current consumption, or go to Dashboards › Usage for expanded utilization metrics.
Apply a reusable agent configuration across enterprise projects to enforce consistent model and tool settings.
📍 In the console (enterprise/SaaS), go to Settings › Agent Profiles, create or select a profile, and assign it to one or more projects.
Adds Agent Profiles on the cloud/SaaS SaaS Software delivered over the internet and billed by subscription, hosted and operated by the vendor so the buyer skips installation and infrastructure management. backend, enabling enterprise teams to define and manage reusable agent configurations. Not checked yet Introduces a Budgets dashboard and expands the Usage Dashboard with new spend and consumption metrics. Not checked yet New budgets and usage monitoring UI surfaces cost and utilization data directly in the product. Not checked yet Surfaces email provider status (SMTP/Resend enabled) in the UI so admins can verify notification configuration at a glance. Not checked yet Passes repository metadata to observability traces, enriching telemetry with source context for agent runs. Not checked yet
5 details in Everything
OpenHands 1.45 adds Agent Profiles, a Budgets dashboard, expanded Usage Dashboard, and email provider status surfacing.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.45.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout cloud-1.45.0
Adds Agent Profiles support on the cloud/SaaS SaaS Software delivered over the internet and billed by subscription, hosted and operated by the vendor so the buyer skips installation and infrastructure management. backend, enabling enterprise-level agent configuration management. Not checked yet Introduces a Budgets dashboard for tracking and enforcing spending limits. Not checked yet Expands the Usage Dashboard with additional monitoring metrics and UI. Not checked yet Surfaces email-enabled status for 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. and Resend email providers in the admin UI. Not checked yet
4 details in Everything
OpenHands 1.45 adds Agent Profiles, a Budgets dashboard, expanded Usage dashboard, and email delivery status surfacing.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.45.0 https://github.com/OpenHands/OpenHands.git
$ git checkout cloud-1.45.0
New Agent Profiles support on the cloud/SaaS SaaS Software delivered over the internet and billed by subscription, hosted and operated by the vendor so the buyer skips installation and infrastructure management. backend lets enterprises define and manage reusable agent configurations. Not checked yet New Budgets dashboard enables teams to set and monitor spending limits across agent usage. Not checked yet Expanded Usage Dashboard provides broader visibility into agent activity and resource consumption. Not checked yet Surfaces email-enabled status for 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. /Resend configurations so admins can confirm delivery is active. Not checked yet
4 details in Everything
OpenHands 1.44.0 adds repository metadata to observability traces for richer debugging context.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.44.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout cloud-1.44.0
OpenHands 1.44.0 adds repository metadata to observability traces for richer pipeline monitoring.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.44.0 https://github.com/OpenHands/OpenHands.git
$ git checkout cloud-1.44.0
Includes repository metadata (e.g., repo name, branch, commit) in observability traces, enabling richer filtering and attribution in tracing backends. Not checked yet
1 detail in Everything
OpenHands cloud gains 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. email service, user login tracking, and a renamed Usage & Monitoring dashboard.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.43.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout cloud-1.43.0
Adds SMTP email service integration for outbound email delivery. Not checked yet Surfaces email-enabled status for both SMTP and Resend email providers in the UI. Not checked yet Tracks user login timestamps for visibility into account activity. Not checked yet Renames the admin dashboard to Usage & Monitoring . Not checked yet
4 details in Everything
OpenHands cloud-1.43.0 adds 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. email service, user login tracking, and a renamed Usage & Monitoring dashboard.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.43.0 https://github.com/OpenHands/OpenHands.git
$ git checkout cloud-1.43.0
└──▷ HOW TO FIND IT
Check whether email delivery (SMTP or Resend) is enabled for your OpenHands Cloud instance — useful before configuring notification workflows.
📍 In the console, go to Usage & Monitoring and look for the email-enabled status indicator showing SMTP or Resend as active.
Adds SMTP email service integration, enabling transactional email delivery alongside the existing Resend provider. Not checked yet New email-enabled status surface shows whether SMTP or Resend email is active in the platform configuration. Not checked yet Tracks user login timestamps, giving admins visibility into last-active data per user. Not checked yet Renames the Admin Dashboard to Usage & Monitoring for clearer intent. Not checked yet
4 details in Everything
OpenHands 1.10.0 adds 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. email service, user login tracking, and a renamed Usage & Monitoring dashboard.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.10.0 https://github.com/OpenHands/OpenHands.git
$ git checkout 1.10.0
└──▷ HOW TO FIND IT
Check whether email delivery is active for your deployment — useful when onboarding users who need invite or notification emails.
📍 In the console, go to Settings › Usage & Monitoring and check the Email Enabled indicator for your configured SMTP or Resend provider.
Adds SMTP email service integration, enabling email delivery alongside the existing Resend provider. Not checked yet Surfaces email-enabled status in the UI for both SMTP and Resend configurations. Not checked yet Tracks user login timestamps, giving admins visibility into account activity. Not checked yet Renames the admin dashboard to Usage & Monitoring for clearer intent. Not checked yet
4 details in Everything
OpenHands 1.42.0 adds a super-admin management API for granting, revoking, and listing admin privileges in enterprise deployments.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.42.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout cloud-1.42.0
New enterprise super-admin management endpoint lets administrators programmatically grant, revoke, and list super-admin roles via API. Not checked yet
1 detail in Everything
OpenHands cloud-1.42.0 adds a super-admin management API for granting, revoking, and listing admin privileges.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.42.0 https://github.com/OpenHands/OpenHands.git
$ git checkout cloud-1.42.0
New enterprise super-admin management endpoint enables granting, revoking, and listing super-admin roles via API. Not checked yet
1 detail in Everything
OpenHands 1.9.1 adds a super-admin management API for granting, revoking, and listing admin privileges.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.9.1 https://github.com/OpenHands/OpenHands.git
$ git checkout 1.9.1
New enterprise/auth super-admin endpoint enables granting, revoking, and listing super-admin roles via API. Not checked yet
1 detail in Everything
OpenHands 1.41 adds tree-sitter tree-sitter A parser-generator library, maintained by the Tree-sitter project, that builds fast, incremental, error-tolerant syntax trees for source code; cyber tools use it to parse and query code without writing a full compiler frontend. semantic chunking, parallel tool calls, time-bounded API keys, and an org admin conversation dashboard.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.41.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout cloud-1.41.0
└──▷ HOW TO FIND IT
Restrict an API key to a specific validity window so it auto-expires after a contractor engagement ends.
📍 In the console, go to Settings › API Keys, create or edit a key, and set the Not Before and Expires At fields to define the active window.
Speed up agent tasks that invoke multiple tools by allowing them to run concurrently instead of sequentially.
📍 In the console, go to Agent Settings and set tool_concurrency_limit to the desired number of parallel tool calls (e.g., 4).
Adds semantic file chunking using tree-sitter AST AST An abstract syntax tree, a structured representation of parsed code. Tools inspect or transform this tree to understand expressions and statements rather than just matching text. parsing for more accurate code understanding during agent tasks. Not checked yet New organization conversation admin dashboard gives org admins visibility into all conversations under their org. Not checked yet Workspace state is now snapshotted at session start and archived before deletion, preserving production workspace history. Not checked yet Parallel tool calls are now configurable via a tool_concurrency_limit agent setting, enabling concurrent tool execution. Not checked yet API keys gain optional active-window controls (not_before and expires_at) for time-bounded access. Not checked yet
+7 more
show less
API keys now support an explicit unbound org scope as a first-class option. Not checked yet New GET /organizations/{id}/me endpoint exposes the caller's permissions within an organization. Not checked yet Enterprise super roles can be assigned via user.role_id with permission fallback for fine-grained access control. Not checked yet Repo and branch metadata is now attached to Laminar trace entries for better observability of agent runs. Not checked yet Dynamic marketplace support added for plugin registration, enabling third-party plugin discovery. Not checked yet SaaS SaaS Software delivered over the internet and billed by subscription, hosted and operated by the vendor so the buyer skips installation and infrastructure management. auth now accepts an api_key cookie as a fallback credential alongside header-based auth. Not checked yet Jira Data Center integration panel now surfaces member-appropriate guidance instead of the admin setup form for non-admin users. Not checked yet
12 details in Everything
OpenHands 1.41 adds API key time windows, parallel tool calls, org admin dashboard, and semantic file chunking.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.41.0 https://github.com/OpenHands/OpenHands.git
$ git checkout cloud-1.41.0
└──▷ TRY IT
Create an API key that is only valid within a specific time window — useful for temporary contractor access or scheduled automation.
copy
$ curl -X POST https://app.openhands.ai/api/api-keys \
-H 'Authorization: Bearer <your-token>' \
-H 'Content-Type: application/json' \
-d '{"name": "ci-bot", "not_before": "2026-08-01T00:00:00Z", "expires_at": "2026-08-31T23:59:59Z"}'
Check what permissions your service account has within an organization before attempting privileged operations.
copy
$ curl -X GET https://app.openhands.ai/api/organizations/<org-id>/me \
-H 'Authorization: Bearer <your-token>'
Enable parallel tool calls in agent settings to let the agent execute multiple tools concurrently and reduce round-trip latency on complex tasks.
📍 In the console, go to Settings › Agent and set tool_concurrency_limit to the desired maximum number of concurrent tool calls (e.g. 4).
Adds semantic file chunking via tree-sitter tree-sitter A parser-generator library, maintained by the Tree-sitter project, that builds fast, incremental, error-tolerant syntax trees for source code; cyber tools use it to parse and query code without writing a full compiler frontend. AST AST An abstract syntax tree, a structured representation of parsed code. Tools inspect or transform this tree to understand expressions and statements rather than just matching text. parsing for more accurate code context in agent sessions. Not checked yet New organization conversation admin dashboard lets org admins review and manage all conversations across their org. Not checked yet Captures production workspace state as an initial snapshot on start and archives it before deletion. Not checked yet Adds super roles via user.role_id with permission fallback for enterprise/auth authorization. Not checked yet Exposes caller permissions on GET /organizations/{id}/me so clients can check what actions they're allowed to take. Not checked yet
+6 more
show less
Adds optional active-window fields (not_before and expires_at) to API keys for time-bounded credentials. Not checked yet Makes unbound org scope an explicit first-class option when creating API keys. Not checked yet Adds parallel tool calls support via tool_concurrency_limit in agent settings to speed up multi-tool workloads. Not checked yet Adds dynamic marketplace support for plugin registration, enabling third-party plugin discovery. Not checked yet Accepts api_key cookie as a fallback credential for SaaS SaaS Software delivered over the internet and billed by subscription, hosted and operated by the vendor so the buyer skips installation and infrastructure management. auth flows. Not checked yet Attaches repo and branch metadata to Laminar traces for better observability of agent runs. Not checked yet
11 details in Everything
OpenHands 1.9.0 adds Azure DevOps resolver, multi-model discovery, 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. config for ACP agents, sub-agent visualizer, and org admin tooling.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.9.0 https://github.com/OpenHands/OpenHands.git
$ git checkout 1.9.0
└──▷ USE IT
Hide personal workspaces on an org-only OHE install so users only see org workspaces.
ini
copy
OH_HIDE_PERSONAL_WORKSPACES=true
OH_DEPLOYMENT_MODE=enterprise
Set a per-deployment resolver label and restrict resolver to enterprise cloud mode.
ini
copy
OH_RESOLVER_LABEL=my-team-resolver
OH_DEPLOYMENT_MODE=enterprise_cloud
Switch the 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. model mid-conversation for a cloud ACP agent via the proxy endpoint.
copy
$ curl -X POST https://<your-openhands-host>/api/acp/switch_acp_model \
-H 'Authorization: Bearer <api_key>' \
-H 'Content-Type: application/json' \
-d '{"conversation_id": "<conversation_id>", "model": "gpt-4o"}'
Adds hide_personal_workspaces flag for org-only OHE installs to enforce org-scoped workspaces. Not checked yet New switch_acp_model proxy endpoint enables model switching for cloud ACP conversations. Not checked yet Adds model dropdown and credential fields to the ACP agent settings page. Not checked yet Adds validation for secret names to ensure valid environment-variable format. Not checked yet Exposes copyable invite links so org invitations work without email delivery. Not checked yet
+33 more
show less
Auto-accepts invitation links on click, removing the confirm/cancel modal. Not checked yet Accepts pending invitations by verified-email match at sign-in. Not checked yet New explicit OH_DEPLOYMENT_MODE flag gates cloud-account CTAs to enterprise cloud deployments. Not checked yet OHE multi-model LLM discovery with BYOK BYOK Bring Your Own Key — a cloud deployment model where the customer supplies and controls the encryption keys rather than the provider, so sensitive data stays under the customer's cryptographic authority. gating: proxy discovery, catalogue union, and updated UX. Not checked yet Allows MCP server configuration for ACP agents. Not checked yet Exposes the Automations menu item to all users. Not checked yet Persists and injects per-user OAuth tokens in Jira DC resolver conversations. Not checked yet Adds DynamicRemoteSandboxSpecService backed by runtime-API warm configs for dynamic sandbox specs. Not checked yet Parallelizes GitHub GraphQL queries in get_suggested_tasks for faster task suggestions. Not checked yet Surfaces a clear error when a custom sandbox image's agent-server SDK version mismatches. Not checked yet New OH_RESOLVER_LABEL env var enables per-deployment resolver macro overrides. Not checked yet Supports Slack attachments in agent conversation context. Not checked yet Adds sub-agent task (TaskToolSet) visualizer to the chat UI. Not checked yet New Full Git history user setting controls how much repository history is provided to the agent. Not checked yet New enterprise admin user-provisioning endpoint for programmatic user management. Not checked yet Uses interrupt endpoint for agent pause, enabling cleaner pause/resume from the UI. Not checked yet Adds default_sandbox_spec_id to user settings for per-user default sandbox selection. Not checked yet Azure DevOps integration: resolver mentions and org-level one-click webhook setup. Not checked yet Jira DC: email-mode auto-link, targeted repo lookup, and picker mentions. Not checked yet Implements semantic file chunking using tree-sitter tree-sitter A parser-generator library, maintained by the Tree-sitter project, that builds fast, incremental, error-tolerant syntax trees for source code; cyber tools use it to parse and query code without writing a full compiler frontend. AST AST An abstract syntax tree, a structured representation of parsed code. Tools inspect or transform this tree to understand expressions and statements rather than just matching text. parsing for improved code context. Not checked yet New organization conversation admin dashboard for org-level conversation oversight. Not checked yet Captures production workspace state with an initial snapshot at start and archive before delete. Not checked yet API keys gain an optional active window via not_before and expires_at fields. Not checked yet Adds tool_concurrency_limit setting to agent configuration for parallel tool calls. Not checked yet Makes unbound org scope an explicit, first-class option for API keys. Not checked yet Adds dynamic marketplace support for plugin registration. Not checked yet SaaS SaaS Software delivered over the internet and billed by subscription, hosted and operated by the vendor so the buyer skips installation and infrastructure management. auth now accepts api_key cookie as a fallback credential. Not checked yet New super-roles via user.role_id with permission fallback for enterprise auth. Not checked yet Exposes caller permissions on GET /organizations/{id}/me. Not checked yet First signed-in user is automatically made the default org owner, tracked via is_default flag. Not checked yet Moves existing users into the default org when first auto-added. Not checked yet Routes unclaimed automation events to the default org on single-org installs. Not checked yet Adds repo/branch metadata to Laminar trace entries for better observability. Not checked yet
38 details in Everything
OpenHands 1.40 adds full Git history cloning, a default sandbox spec setting, Azure DevOps webhook setup, and richer Jira DC integration.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.40.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout cloud-1.40.0
└──▷ HOW TO FIND IT
Pin a specific sandbox environment for all your agent runs so you don't have to reconfigure it per session.
📍 In the console, go to Settings › User Settings and set the Default Sandbox Spec to your preferred environment via the default_sandbox_spec_id field.
Enable full Git history cloning when your agents need to analyze commit history or run git-log-based tools.
📍 In the console, go to Settings › User Settings and enable the Full Git History option.
Set up org-level Azure DevOps webhooks in one click to automatically trigger OpenHands resolvers on mentions.
📍 In the console, go to Settings › Integrations › Azure DevOps and use the one-click webhook setup under your organization.
Adds a user setting to enable full Git history cloning (instead of shallow clones) for agents that need deep commit context. Not checked yet Adds default_sandbox_spec_id to user settings, letting users pin a preferred sandbox environment across sessions. Not checked yet Adds an enterprise admin endpoint for programmatic user provisioning. Not checked yet Adds Azure DevOps resolver mentions and org-level one-click webhook setup, simplifying CI/CD trigger configuration. Not checked yet Enhances Jira Data Center integration with email-mode auto-linking, targeted repository lookup, and picker mentions. Not checked yet
+1 more
show less
Introduces a dedicated interrupt endpoint powering the agent pause UI, enabling more reliable mid-task agent control. Not checked yet
6 details in Everything
OpenHands 1.40 adds full Git history cloning, Azure DevOps webhook setup, Jira DC enhancements, and a default sandbox spec setting.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.40.0 https://github.com/OpenHands/OpenHands.git
$ git checkout cloud-1.40.0
└──▷ HOW TO FIND IT
Set a default sandbox spec so every new conversation automatically uses your preferred compute environment — no manual selection required.
📍 In the console, go to Settings › User Settings and set Default Sandbox Spec to your preferred spec, then save.
Enable full Git history cloning when you need access to complete commit logs and blame history inside the agent sandbox.
📍 In the console, go to Settings › User Settings, enable Full Git History, then save.
Set up Azure DevOps org-level webhook in one click so any resolver mention automatically triggers an OpenHands task.
📍 In the console, go to Settings › Integrations › Azure DevOps and click Set Up Webhook at the org level.
Adds Full Git History user setting to control whether repositories are cloned with complete commit history instead of a shallow clone. Not checked yet New enterprise admin endpoint for user provisioning, enabling programmatic creation and management of org users via API. Not checked yet Adds default_sandbox_spec_id to user settings, letting users designate a preferred sandbox environment spec for new conversations. Not checked yet Azure DevOps integration gains resolver mentions support and org-level one-click webhook setup, simplifying automated task triggering. Not checked yet Jira Data Center integration adds email-mode auto-linking, targeted repository lookup, and picker-based @mentions for richer issue-to-repo workflows. Not checked yet
+1 more
show less
Agent pause UI now uses a dedicated interrupt endpoint, enabling more reliable mid-task agent pausing. Not checked yet
6 details in Everything
OpenHands 1.39.0 adds ACP model switching, 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. server config for ACP agents, Slack attachment context, multi-model 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. discovery, copyable invite links, and a sub-agent task visualizer.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.39.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout cloud-1.39.0
└──▷ TRY IT
Pin a custom resolver label per deployment so automation events are routed to the correct resolver macro.
copy
$ OH_RESOLVER_LABEL=my-org-resolver
Copy and share an org invite link directly from the console so new members can join without waiting for an email.
📍 In the console, go to Settings › Organization › Members and click Copy invite link to share a URL that auto-accepts on click.
Adds switch_acp_model proxy endpoint enabling model switching for cloud ACP conversations. Not checked yet Adds model dropdown and credential fields to the agent settings page for ACP agents. Not checked yet Supports MCP server configuration for ACP agents. Not checked yet Enables multi-model LLM discovery and BYOK BYOK Bring Your Own Key — a cloud deployment model where the customer supplies and controls the encryption keys rather than the provider, so sensitive data stays under the customer's cryptographic authority. gating for OpenHands Enterprise (OHE) installs, including proxy discovery and catalogue union. Not checked yet Adds hide_personal_workspaces flag to restrict OHE installs to org-only workspaces. Not checked yet
+16 more
show less
Adds explicit OH_DEPLOYMENT_MODE flag to control deployment mode and gate cloud-account CTAs to enterprise cloud. Not checked yet Adds OH_RESOLVER_LABEL env var to configure a per-deployment resolver macro. Not checked yet Exposes copyable invite links so org invitations work without email delivery. Not checked yet Auto-accepts invitation links on click, removing the confirm/cancel modal. Not checked yet Accepts pending invitations by verified-email match at sign-in. Not checked yet Adds validation for secret names to ensure valid environment variable format. Not checked yet Surfaces Slack attachments as context in agent conversations. Not checked yet Adds a sub-agent task (TaskToolSet) visualizer to the chat UI. Not checked yet Adds DynamicRemoteSandboxSpecService backed by runtime-API warm configs for dynamic sandbox provisioning. Not checked yet Parallelizes GitHub GraphQL queries in get_suggested_tasks for faster task suggestions. Not checked yet Surfaces a clear error in the UI when a custom sandbox image's agent-server SDK version mismatches. Not checked yet Persists and injects per-user OAuth tokens in Jira Data Center resolver conversations. Not checked yet Exposes the Automations menu item to all users. Not checked yet Routes unclaimed automation events to the default org on single-org installs. Not checked yet Moves existing users into the default org when first auto-added, and makes the first signed-in user the default org owner (tracked via is_default flag). Not checked yet Encodes agent harness agent harness The software around an AI agent that manages its tools, execution loop, state and limits. It determines how the model's proposed actions are run and checked. in the conversation chip icon and surfaces the model name in chip text. Not checked yet
21 details in Everything
OpenHands 1.39.0 adds multi-model BYOK BYOK Bring Your Own Key — a cloud deployment model where the customer supplies and controls the encryption keys rather than the provider, so sensitive data stays under the customer's cryptographic authority. discovery, 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. server config for ACP agents, Slack attachment context, and org-management improvements.
└──▷ GET THIS VERSION
copy
$ git clone --branch cloud-1.39.0 https://github.com/OpenHands/OpenHands.git
$ git checkout cloud-1.39.0
└──▷ HOW TO FIND IT
Hide personal workspaces for an org-only OpenHands Enterprise deployment so users only see org-scoped workspaces.
📍 In the OpenHands Enterprise admin console, go to Organization Settings and enable the hide_personal_workspaces flag.
Share an invite link directly with a new org member when email delivery is unavailable.
📍 In the OpenHands console, go to Organization Settings › Members › Invite Members, then copy the generated invite link and share it directly with the invitee.
Adds hide_personal_workspaces flag for org-only OpenHands Enterprise installs, restricting users to org workspaces only. Not checked yet Introduces a switch_acp_model proxy endpoint enabling model switching within cloud ACP conversations. Not checked yet Adds model dropdown and credential fields to the agent settings page for ACP agents. Not checked yet Supports MCP server configuration for ACP agents, enabling custom tool server connections. Not checked yet Enables multi-model 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. discovery and BYOK gating in OpenHands Enterprise, unifying proxy discovery and the model catalogue. Not checked yet
+18 more
show less
Adds validation for secret names to ensure they conform to valid environment variable format. Not checked yet Exposes copyable invite links so org invitations work without requiring email delivery. Not checked yet Auto-accepts invitation links on click, removing the confirm/cancel modal step. Not checked yet Accepts pending invitations by verified-email match at sign-in for frictionless onboarding. Not checked yet Routes unclaimed automation events to the default org on single-org installs. Not checked yet Makes the first signed-in user the default org owner and tracks orgs via an is_default flag. Not checked yet Moves existing users into the default org automatically when first auto-added. Not checked yet Adds explicit OH_DEPLOYMENT_MODE flag to gate cloud-account CTAs to enterprise cloud deployments. Not checked yet Supports per-deployment resolver macro customization via OH_RESOLVER_LABEL environment variable. Not checked yet Adds Slack attachment support in agent context so agents can read files and images sent via Slack. Not checked yet Adds a sub-agent task (TaskToolSet) visualizer to the chat UI for monitoring parallel agent workloads. Not checked yet Parallelizes GitHub GraphQL queries in get_suggested_tasks for faster task suggestion loading. Not checked yet Exposes the Automations menu item to all users. Not checked yet Persists and injects per-user OAuth tokens in Jira Data Center resolver conversations. Not checked yet Shares linked OAuth tokens with eligible Jira DC conversations. Not checked yet Surfaces a clear error in the UI when a custom sandbox image's agent-server SDK version mismatches. Not checked yet Encodes harness in the conversation chip icon and surfaces model name in chip text. Not checked yet Adds DynamicRemoteSandboxSpecService backed by runtime-API warm configs for dynamic sandbox provisioning. Not checked yet
23 details in Everything
OpenHands 1.8.0 adds 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. profiles, sub-agent delegation, sandbox grouping strategy, and an ACP agent UI.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.8.0 https://github.com/OpenHands/OpenHands.git
$ git checkout 1.8.0
└──▷ HOW TO FIND IT
Switch between saved LLM configurations (e.g., a fast model for drafting and a powerful model for review) without re-entering credentials each time.
📍 In the console, go to Settings › LLM and create or select an LLM profile with your preferred model, API key, and parameters.
Control how sandboxes are shared or isolated across conversations to optimize resource usage or enforce separation.
📍 In the console, go to Settings › Sandbox and select the desired sandbox grouping strategy.
Adds LLM profiles, letting users save and switch between named LLM configurations. Not checked yet Enables sub-agent delegation so the primary agent can spawn and coordinate specialized sub-agents for complex tasks. Not checked yet Adds selectable sandbox grouping strategies, giving users control over how conversation sandboxes are organized and reused. Not checked yet Introduces a minimal generic ACP (Agent Communication Protocol) agent UI for interacting with ACP-compatible agents. Not checked yet
4 details in Everything
OpenHands 1.6.0 adds hooks support, /clear and /new chat commands, toggleable global skills, and Catalan UI localization.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.6.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 1.6.0
└──▷ HOW TO FIND IT
Reset chat context mid-session without losing your running sandbox state — useful when a conversation has drifted and you want a clean prompt history.
📍 In the chat input, type /clear and press Enter. A new conversation opens, inheriting the current sandbox and configuration.
Adds hooks support, enabling extensible event-driven integrations with the OpenHands runtime. Not checked yet New /clear slash command starts a fresh conversation that inherits the current sandbox and configuration, preserving runtime state while resetting chat history. Not checked yet New /new slash command is now surfaced in the slash command menu for faster access to new conversations. Not checked yet Adds ability to enable or disable default global skills, giving operators control over which built-in capabilities are active. Not checked yet Adds a copy button to code blocks in the chat UI for easier extraction of generated code. Not checked yet
+1 more
show less
Adds Catalan language support to the UI. Not checked yet
6 details in Everything
OpenHands 1.5.0 adds a planning agent, real-time Task List panel, slash command menu, Bitbucket Datacenter support, and six new models.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.5.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 1.5.0
└──▷ HOW TO FIND IT
Surface available agent skills mid-conversation without leaving the chat input.
📍 In the chat input, type / to open the slash command menu and select an available agent skill or command.
Monitor the agent's progress across multi-step tasks in real time.
📍 In the right panel, click the Task List tab to view the agent's current tasks and their live status.
Adds a planning agent for structured, multi-step task execution. Not checked yet New Task List tab in the right panel shows the agent's current task list with real-time status updates. Not checked yet New slash command menu in the chat input surfaces available agent skills and commands when typing /. Not checked yet Enables attaching or changing the Git repository for existing conversations. Not checked yet Adds Bitbucket Datacenter as a supported Git provider. Not checked yet
+1 more
show less
Adds support for Claude Opus 4.6, Claude Sonnet 4.6, GLM-4.7, GLM-5, Kimi-K2.5, Qwen3-Coder-Next, and Gemini 3.1 Pro Preview models. Not checked yet
6 details in Everything
OpenHands 1.4.0 adds MiniMax-M2.5 model support.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.4.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 1.4.0
Adds MiniMax-M2.5 as a supported 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. backend. Not checked yet
1 detail in Everything
OpenHands 1.3.0 adds 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. support for remote browser access and host networking mode for reverse proxy setups.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.3.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 1.3.0
Adds CORS support to the Docker sandbox service, enabling remote browser access by setting WEB_HOST and SANDBOX_CONTAINER_URL_PATTERN. Not checked yet Adds host networking mode via OH_SANDBOX_USE_HOST_NETWORK=true, allowing reverse proxies to reach user-launched applications on any port beyond the predefined worker ports. Not checked yet
2 details in Everything
OpenHands 1.2.0 adds a chat readiness status indicator and doubles the condenser 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. default.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.2.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 1.2.0
Adds a status indicator above the chat UI that shows when a conversation is ready to receive input. Not checked yet Doubles the condenser default max_size from 120 to 240, allowing larger context to be retained before summarization. Not checked yet
2 details in Everything
OpenHands 1.1.0 adds OAuth 2.0 OAuth 2.0 An open authorization framework (IETF RFC 6749) that lets a service grant limited API access to a third party without sharing credentials, giving cyber tools a standard way to authenticate against external platforms. Device Flow for CLI auth, Forgejo integration, and conversation export.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.1.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 1.1.0
└──▷ HOW TO FIND IT
Authenticate the OpenHands CLI from a headless server or remote shell without needing a browser on the same machine.
📍 In the OpenHands CLI, select Login and follow the Device Flow prompt: visit the displayed URL on any device and enter the code to authorize the session.
Save a full agent conversation for audit, review, or sharing — useful after a complex remediation session.
📍 In the OpenHands UI, open the conversation panel, locate the session, and click Export Conversation to download the conversation as a file.
Adds OAuth 2 OAuth 2 An open authorization framework (RFC 6749) that lets a service grant limited access to a user's resources on another service without exposing credentials, giving cyber tools a standard way to authenticate against third-party APIs. .0 Device Flow for CLI authentication, enabling headless and browser-less login workflows. Not checked yet Adds Forgejo integration, extending repository provider support to the self-hosted Forgejo platform. Not checked yet Adds an Export Conversation button in the conversation panel to save full conversation history. Not checked yet Adds a Refresh button on the Changes tab for on-demand file-diff updates without reloading. Not checked yet
4 details in Everything
OpenHands 1.0.0 adds Azure DevOps integration, a new task tracker UI, and 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. for Gemini 2.5 Pro.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.0.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 1.0.0
Adds Azure DevOps integration support for repository and pipeline connectivity. Not checked yet New task tracker interface provides structured visibility into agent task progress. Not checked yet Chat messages are now queued during runtime connection, enabling smoother interaction during agent startup. Not checked yet
3 details in Everything
OpenHands 0.62.0 adds conversation tab pinning and a Documentation link in the profile menu.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.62.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.62.0
└──▷ HOW TO FIND IT
Keep a frequently referenced conversation always accessible by pinning it.
📍 In the OpenHands UI, right-click or use the tab options on a conversation tab and select Pin to pin it; select Unpin to remove the pin.
Enables pinning and unpinning of conversation tabs for quick access to important sessions. Not checked yet Adds a Documentation link directly in the profile menu for faster access to reference material. Not checked yet
2 details in Everything
OpenHands 0.61.0 makes the terminal tab read-only and removes the Jupyter tab from the UI.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.61.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.61.0
Terminal tab is now read-only, preventing direct user input into the agent's terminal session. Not checked yet Jupyter tab has been removed from the UI. Not checked yet
2 details in Everything
└──▷ BREAKING ON UPGRADE
! The Jupyter tab has been removed from the UI; any workflows relying on it will no longer have access to it.
OpenHands 0.60.0 adds Clarifai provider support and automatic sandbox lifecycle management to curb resource exhaustion.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.60.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.60.0
Adds Clarifai as a supported 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. Not checked yet Adds automatic sandbox lifecycle management that limits the number of concurrently running sandboxes to prevent resource exhaustion. Not checked yet Simplifies the security analyzer confirmation UI by replacing two reject options with a single Reject option that prompts for an optional reason. Not checked yet
3 details in Everything
└──▷ BREAKING ON UPGRADE
! The V0 CLI has been removed and is officially unsupported; any setup relying on the V0 CLI will break on upgrade.
OpenHands 0.59.0 adds Lemonade as a supported 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.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.59.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.59.0
OpenHands 1.0.0-cli ships standalone binaries, faster startup, a refreshed UI, and MCP OAuth MCP OAuth An authentication layer for the Model Context Protocol that lets MCP clients obtain delegated access tokens so cyber tools can call MCP servers on a user's behalf without handling raw credentials. support.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.0.0-cli https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 1.0.0-cli
Adds multi-platform standalone executable binaries so the CLI runs without any additional setup or runtime install. Not checked yet Faster CLI startup when running via uv, pip, and similar tools. Not checked yet New and refreshed UI for the CLI experience. Not checked yet Simplified JSON 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. configuration with added MCP OAuth support. Not checked yet
4 details in Everything
OpenHands 0.58.0 adds configurable SHTTP 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. server timeouts and claude-sonnet-4-5 model support.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.58.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.58.0
Adds configurable timeout settings for SHTTP MCP servers, preventing tool calls from hanging indefinitely. Not checked yet Adds support for the claude-sonnet-4-5 model. Not checked yet Converts the settings panel to a vertical sidebar layout. Not checked yet
3 details in Everything
OpenHands 0.57.0 adds a redesigned UI and Bring Your Own Key (BYOK BYOK Bring Your Own Key — a cloud deployment model where the customer supplies and controls the encryption keys rather than the provider, so sensitive data stays under the customer's cryptographic authority. ) support on Cloud.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.57.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.57.0
└──▷ HOW TO FIND IT
Use your own 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 on OpenHands Cloud instead of the platform's default billing.
📍 In the OpenHands Cloud console, go to Settings and enter your own API key under the Bring Your Own Key (BYOK) section.
New UI for landing and conversation pages improves the agent interaction experience. Not checked yet Adds Bring Your Own Key (BYOK) support on OpenHands Cloud, letting users supply their own LLM API keys. Not checked yet
2 details in Everything
OpenHands 0.56.0 adds AGENTS.md support, microagent management, and multi-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. routing within agent trajectories.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.56.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.56.0
Supports AGENTS.md files in microagents, enabling repo-level agent behavior configuration via a standardized file. Not checked yet New microagent management feature for organizing and controlling microagents. Not checked yet Enables routing between different LLMs LLMs 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. within a single agent trajectory, allowing dynamic model selection per task step. Not checked yet
3 details in Everything
OpenHands 0.55.0 adds an 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. risk analyzer and configurable condenser history size limit.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.55.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.55.0
Adds a configurable max history size setting for the condenser, giving operators control over 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. usage. Not checked yet Improves performance and UX when working with repositories containing many branches. Not checked yet
2 details in Everything
OpenHands 0.54 adds task tracking with planning mode, 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. on Cloud, and an improved MCP config UI.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.54.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.54.0
└──▷ HOW TO FIND IT
Connect an MCP server on OpenHands Cloud to extend the agent with custom tools.
📍 In the OpenHands Cloud console, go to Settings › MCP and use the Add, Edit, or Delete controls to manage your MCP server configurations.
Adds a task tracking tool and planning mode to help agents manage long-horizon, multi-step tasks. Not checked yet Enables Model Context Protocol Model Context Protocol 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. (MCP) support on OpenHands Cloud. Not checked yet Improves the MCP configuration UI with add, edit, and delete functionality. Not checked yet Adds a new user feedback reason: The agent didn't finish the job . Not checked yet
4 details in Everything
OpenHands 0.53.0 adds named Docker volume binding, configurable git commit identity, and partial CLI settings updates.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.53.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.53.0
Supports binding named Docker volumes into the runtime for persistent or shared storage across agent sessions. Not checked yet Enables specifying the user identity (name/email) OpenHands commits as when pushing code changes. Not checked yet Allows partial modification of CLI settings without needing to restate the full configuration. Not checked yet
3 details in Everything
OpenHands 0.52.0 adds GPT-5, Claude Opus 4.1, Qwen-3-Coder-480B, CLI UI launch, and overlay mount support.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.52.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.52.0
└──▷ HOW TO FIND IT
Toggle solvability analysis on or off for the GitHub issue resolver in OpenHands Cloud to control triage depth.
📍 In OpenHands Cloud, go to the Resolver settings and use the solvability analysis switch to enable or disable additional solvability analysis.
Adds qwen-3-coder-480b to the OpenHands provider model roster. Not checked yet Adds gpt-5-2025-08-07 to the OpenHands provider model roster. Not checked yet Adds claude-opus-4-1-20250805 as a supported model. Not checked yet Adds search API key settings configuration to the CLI. Not checked yet Supports launching the full OpenHands UI directly from the CLI. Not checked yet
+6 more
show less
Supports overlay mounts in the Docker runtime for more flexible workspace setups. Not checked yet Adds a toggle to enable or disable solvability analysis in the resolver on OpenHands Cloud. Not checked yet Includes branch information in repository context provided to agents. Not checked yet Adds auto-pagination with infinite scroll for the conversation list. Not checked yet Expands conversation history limit from 20 to 100 entries. Not checked yet Adds proper markdown rendering for agent messages and final thoughts in the CLI interface. Not checked yet
11 details in Everything
OpenHands 0.51 adds 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. support for the CLI, multi-repo git change detection, and smarter resolver summaries.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.51.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.51.0
Adds MCP (Model Context Protocol Model Context Protocol 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. ) support to the CLI, enabling tool-server integrations previously only available in the UI. Not checked yet Optimizes git change detection in the Changes tab with performance improvements and multi-repository support. Not checked yet Improves resolver summary generation by focusing each summary only on new changes since the last one, reducing repetition across consecutive summaries. Not checked yet
3 details in Everything
OpenHands 0.50 adds Moonshot AI Kimi-K2 model support and new CLI confirmation dialog options
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.50.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.50.0
Adds Moonshot AI Kimi-K2 (kimi-k2-0711-preview) as a recommended and natively supported model via the OpenHands provider. Not checked yet Adds new options to CLI confirmation dialogs, giving practitioners more control over interactive approval flows. Not checked yet Persists alias choices in the CLI across sessions, reducing repetitive configuration. Not checked yet Improves 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. settings UI layout and clarity for easier Model Context Protocol Model Context Protocol 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. configuration. Not checked yet
4 details in Everything
OpenHands 0.49 adds CLI/VSCode integration, a new Memory UI, OpenHands Cloud 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 conversation management improvements.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.49.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.49.0
└──▷ TRY IT
Check which config file OpenHands is loading without digging through docs — useful when managing multiple environments.
Adds CLI/VSCode integration, enabling OpenHands to be used directly from the command line and VS Code. Not checked yet Introduces an OpenHands Cloud LLM provider, letting users route model calls through OpenHands Cloud. Not checked yet New Memory UI feature surfaces agent memory state in the interface. Not checked yet Conversation cards now display branch name and git provider at a glance. Not checked yet Suggested tasks are now split and organized by git provider. Not checked yet
+3 more
show less
Users can edit a conversation's title directly in the UI. Not checked yet CLI alias setup for first-time users: on first run, users are offered the option to create openhands and oh shell aliases for faster launch. Not checked yet CLI /settings command now displays the active configuration file path. Not checked yet
8 details in Everything
OpenHands 0.48 adds user-directory microagents, .cursorrules support, conversation stop control, and setup.sh event visibility.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.48.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.48.0
Supports loading microagents from the user directory (~/.openhands/microagents/) for per-user agent customization without touching project repos. Not checked yet Supports .cursorrules files placed in the project root directory, letting teams reuse existing Cursor editor rule sets. Not checked yet Adds ability to stop in-progress conversations from the UI. Not checked yet Surfaces setup.sh script execution in the event stream so users can observe when and how the setup script runs. Not checked yet
4 details in Everything
└──▷ BREAKING ON UPGRADE
! The Jupyter plugin is now disabled by default in the CLI runtime; setups that relied on Jupyter being active without explicit configuration will no longer have it enabled.
OpenHands 0.47 adds GitLab microagent config directory support and smarter terminal output truncation.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.47.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.47.0
└──▷ HOW TO FIND IT
Store org-level microagents in a GitLab repo using the supported alternative directory path so OpenHands can discover them.
📍 In your GitLab group, create a repository named openhands-config and place your microagent .md files under the microagents/ folder — OpenHands will automatically load them per the updated org microagent discovery logic.
Adds new feedback options for users within the UI. Not checked yet Supports openhands-config as an alternative GitLab directory for hosting org-level microagents. Not checked yet Changes terminal truncation to trim the middle of long outputs instead of the suffix, preserving both start and end context. Not checked yet
3 details in Everything
OpenHands 0.46 adds custom model names, vi-mode CLI keybindings, configurable safety settings, and system prompt overrides.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.46.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.46.0
└──▷ TRY IT
Enable vi-style j/k navigation in CLI confirmation menus for keyboard-driven workflows.
copy
$ CLI_VI_MODE=true python -m openhands.core.cli
Supports custom/arbitrary model names during CLI model selection, beyond the preset list. Not checked yet Adds optional vi-style keybindings (j/k navigation) for CLI confirmation prompts, enabled via the CLI_VI_MODE environment variable. Not checked yet Adds customizable safety settings for Mistral AI and Gemini models. Not checked yet Enables overriding the hardcoded system prompt for full prompt control. Not checked yet Allows users to submit feedback on agent performance even when the agent hit an error or is awaiting user input. Not checked yet
5 details in Everything
OpenHands 0.45 adds Kubernetes runtime, Bitbucket integration, file/image uploads, per-conversation budgets, and Japanese UI.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.45.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.45.0
Adds Kubernetes Runtime support as a new sandbox execution backend. Not checked yet Supports file and image uploads directly within conversations. Not checked yet Adds Bitbucket integration for local usage alongside existing GitHub support. Not checked yet Enables per-conversation budget configuration to cap 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. spending. Not checked yet Adds Japanese language support to the UI. Not checked yet
+1 more
show less
Increases max iterations per task from 250 to 500, enabling more complex autonomous workflows. Not checked yet
6 details in Everything
└──▷ BREAKING ON UPGRADE
! The local configuration directory has moved from ~/.openhands-state to ~/.openhands; existing state stored in the old path will not be picked up automatically after upgrading.
OpenHands 0.44 adds draft PR/MR control via Git 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. , Slack integration in settings, and SWE-bench-Live evaluation support.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.44.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.44.0
└──▷ HOW TO FIND IT
Install the Slack integration for your Cloud OpenHands workspace without leaving the product.
📍 In the Cloud OpenHands console, go to Settings › Integrations and follow the prompts to install the Slack integration.
Agents can now control whether PRs/MRs opened via Git MCP are created as drafts or ready-for-review. Not checked yet Adds Slack integration installable directly from the Integrations tab in the Cloud OpenHands Settings page. Not checked yet Supports evaluation runs on SWE-bench-Live for more realistic, up-to-date issue-resolving benchmarks. Not checked yet Dev container networking now works without host network mode, improving security and portability of the dev setup. Not checked yet
4 details in Everything
OpenHands 0.43 adds Slack integration Beta , a Microagents UI, GitLab lower-scoped token support, and a code review microagent example.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.43.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.43.0
└──▷ HOW TO FIND IT
Browse available microagents mid-conversation to understand what specialized knowledge is loaded before asking a domain-specific question.
📍 In the conversation UI, open the context menu and select Microagents to view the list of active specialized microagents for the current session.
New Slack integration Beta with Cloud OpenHands for triggering and interacting with agents via Slack. Not checked yet New Microagents UI in the conversation context menu lets users see what specialized knowledge microagents are available in their current conversation. Not checked yet New example code review microagent provides a ready-made template for automated code review workflows. Not checked yet Supports lower-scoped GitLab tokens, enabling more granular permission control when connecting GitLab repositories. Not checked yet Adds JSON serialization for array and object parameters when converting tools, broadening tool-call compatibility. Not checked yet
5 details in Everything
└──▷ BREAKING ON UPGRADE
! The CLI default provider is changed from openai to anthropic; existing CLI users relying on the default will now target Anthropic instead of OpenAI.! CLI settings are now saved directly under ~/.openhands instead of the previous location; existing CLI config files stored elsewhere will no longer be read automatically.
OpenHands 0.42.0 adds API endpoints to control conversation lifecycle and lets users update their email address.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.42.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.42.0
New API endpoints to explicitly start and stop conversations programmatically. Not checked yet New VSCODE_IN_NEW_TAB feature flag to open the VS Code editor in a new browser tab. Not checked yet
2 details in Everything
OpenHands 0.41.0 adds Cloud GitLab Resolver, streamable HTTP 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. support, and native Windows runtime.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.41.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.41.0
Adds Cloud GitLab Resolver, enabling automated issue/PR resolution against cloud-hosted GitLab repositories. Not checked yet Implements streamable HTTP MCP transport, supporting MCP servers served over HTTP streams. Not checked yet Adds native Windows support without requiring WSL WSL Windows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. , broadening local deployment options. Not checked yet Improves CLI mode setup flow to guide users through configuration when settings are missing. Not checked yet
4 details in Everything
OpenHands 0.40.0 adds native PR/MR creation via 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. , LocAgent integration, and a first-class search API.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.40.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.40.0
Enables native PR/MR opening on GitHub and GitLab via in-house MCP integration. Not checked yet Incorporates LocAgent into OpenHands for improved code localization capabilities. Not checked yet Adds first-class search API support to OpenHands. Not checked yet Adds Interactive SWE-Bench benchmark for evaluating agent performance. Not checked yet Improves MCP tool usage visualization in the UI. Not checked yet
5 details in Everything
OpenHands 0.39.0 adds UI-based custom secret management and nested runtime API support for scalable agent deployments.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.39.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.39.0
└──▷ HOW TO FIND IT
Store a custom API key or credential so the agent can securely access it during task execution — no code change required.
📍 In the console, go to Settings › Secrets and add a new custom secret (name + value) to make it available to the agent at runtime.
Adds ability to save custom secrets for agents directly through the UI settings panel. Not checked yet Adds API support for nested runtimes, enabling more flexible and scalable agent deployments. Not checked yet
2 details in Everything
└──▷ BREAKING ON UPGRADE
! Only volumes explicitly mounted to /workspace will be treated as workspace mounts; other volume mounts no longer receive workspace treatment on upgrade.
OpenHands 0.38.0 adds Windows runtime, 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. config UI, self-hosted Git support, and user/org microagents.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.38.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.38.0
└──▷ HOW TO FIND IT
Configure MCP servers for the agent via the settings UI.
📍 In the OpenHands UI, go to Settings › MCP and add your MCP server configurations.
Adds self-hosted GitLab and enterprise GitHub host settings for local installations. Not checked yet Supports user-defined runners for the resolver, enabling custom execution environments. Not checked yet Adds Windows local runtime support with PowerShell. Not checked yet Enables the agent to save browser screenshots as image files during sessions. Not checked yet Supports loading custom Agent implementations from arbitrary Python packages via config.toml. Not checked yet
+3 more
show less
Adds MCP (Model Context Protocol Model Context Protocol 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. ) configuration directly in the settings UI. Not checked yet Supports user- and org-level microagents for shared, reusable agent configurations. Not checked yet Simplifies workspace mounting via the SANDBOX_VOLUMES configuration. Not checked yet
8 details in Everything
OpenHands 0.37.0 adds branch picker for repositories and per-user configurable conversation starters.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.37.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.37.0
└──▷ HOW TO FIND IT
Target a specific branch when starting work on a repository from the homepage, without manually switching branches later.
📍 In the console, go to the Homepage, select your repository, then use the branch picker to choose the desired branch before starting a conversation.
Disable proactive conversation starters for Cloud Resolver if you prefer a quieter, prompt-only experience.
📍 In the console, go to Account Settings and toggle Proactive conversation starters off.
Adds a branch picker for repository selection on the homepage. Not checked yet Enables per-user configuration of proactive conversation starters for Cloud Resolver, togglable in account settings (enabled by default). Not checked yet Adds helpful tips to the Changes tab zero state. Not checked yet
3 details in Everything
OpenHands 0.36.0 adds CLI agent pause/resume, documentation search, and raises the conversation panel limit to 20.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.36.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.36.0
Adds pause and resume functionality for agents running in the CLI. Not checked yet Adds search support for documentation within the product. Not checked yet Separates settings into multiple distinct sections in the UI. Not checked yet Increases the conversation panel limit from 9 to 20. Not checked yet
4 details in Everything
OpenHands 0.35 adds API key management UI, GitLab suggested tasks, a VS Code tab, and new CLI commands.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.35.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.35.0
└──▷ TRY IT
Quickly start a new session, check status, or open settings without leaving the CLI.
copy
$ /new
/status
/settings
Manage API keys for programmatic access to OpenHands Cloud from the settings page.
📍 In the OpenHands Cloud console, go to Settings › API Keys to create or revoke keys.
Supports viewing and launching suggested tasks for GitLab repositories directly from the homepage. Not checked yet Replaces the Workspace tab with a VS Code tab for in-browser code editing. Not checked yet Enables /new, /status, and /settings commands in the CLI. Not checked yet Sorts GitHub repos in OpenHands Cloud by most recently pushed, making active repos easier to find. Not checked yet Simplifies microagent authoring: names in frontmatter are no longer required — only triggers matter for dynamically loaded microagents. Not checked yet
5 details in Everything
OpenHands 0.34 adds a Changes tab for reviewing agent modifications, custom sandboxes for the resolver, and tool disabling controls.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.34.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.34.0
└──▷ HOW TO FIND IT
When reviewing what an agent changed in a session, open the Changes tab to see a diff of all modified files before accepting or rolling back.
📍 In the OpenHands UI, open an active or completed session and click the Changes tab to view a diff of all files modified by the agent.
New Changes tab in the UI surfaces all file modifications made by OpenHands during a session, making agent-driven diffs reviewable at a glance. Not checked yet Adds ability to disable specific default tools, giving operators fine-grained control over which capabilities the agent can invoke. Not checked yet Supports custom sandbox images for the resolver, enabling reproducible, project-specific environments when resolving GitHub issues. Not checked yet Resolver now names branches with the openhands/ prefix, establishing a consistent convention for identifying agent-created branches. Not checked yet Updated home screen layout improves navigation and workspace entry points. Not checked yet
5 details in Everything
└──▷ BREAKING ON UPGRADE
! Resolver branches are now created with the openhands/ prefix — any automation or filters matching previous branch name patterns must be updated.
OpenHands 0.33 adds native 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. server support, GitLab OAuth/PAT PAT A personal access token, a credential used by scripts and integrations instead of a user's password. Its permissions and expiry depend on the service issuing it. integration, GitHub Enterprise resolver, and interactive CLI commands.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.33.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.33.0
Supports MCP servers natively via CodeActAgent, enabling model context protocol tool use tool use 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. within agents. Not checked yet Adds GitLab API integration with both OAuth and PAT token authentication. Not checked yet Enables PR type specification when using the GitHub Action resolver. Not checked yet Supports GitHub Enterprise and on-premises GitLab in the GitHub Action resolver. Not checked yet Adds interactive CLI commands /init, /help, and /exit with an enhanced startup experience. Not checked yet
+2 more
show less
Displays agent think actions as collapsible elements in the conversation view. Not checked yet Improves SWE-bench_Verified single-pass baseline results to 59.6%. Not checked yet
7 details in Everything
OpenHands 0.32 adds microagent context visibility, PR template awareness, and enables context condensation by default.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.32.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.32.0
Displays microagent context directly in the UI so practitioners can see which microagent knowledge is active during a session. Not checked yet Shows edited file paths in the conversation headline for at-a-glance change tracking. Not checked yet Enables context condensation by default, automatically compressing long conversation histories to stay within model context limits. Not checked yet Expands GitLab repository listing to include repos where the user has membership but is not the owner. Not checked yet Supports microagent files without a required header, reducing friction when authoring custom microagents. Not checked yet
+3 more
show less
Adds a user-friendly UI message for content policy violation errors. Not checked yet Prompts the agent to follow existing PR templates when creating new pull requests. Not checked yet Improves event retrieval performance for conversations with a large number of events. Not checked yet
8 details in Everything
└──▷ BREAKING ON UPGRADE
! Condensation is now enabled by default; existing deployments that relied on full, uncompressed conversation history will have condensation applied automatically on upgrade.
OpenHands 0.31.0 adds vision-based browsing with PDF/image support, OpenHands LM integration, and multi-provider Git configuration.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.31.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.31.0
└──▷ HOW TO FIND IT
Configure both GitHub and GitLab as Git providers at the same time for local OpenHands, so agents can operate across repositories on either platform.
📍 In the OpenHands settings UI, go to Settings › Git Providers and add credentials for both GitHub and GitLab simultaneously.
Enables vision-based browsing by default, allowing OpenHands to see and interpret PDFs and images rendered in the browser. Not checked yet Adds support for OpenHands LM as a new 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 option. Not checked yet Supports configuring multiple Git providers simultaneously (e.g., GitHub + GitLab) in local OpenHands deployments. Not checked yet Generates LLM-based natural-language titles for conversations automatically. Not checked yet
4 details in Everything
OpenHands 0.30.1 adds trajectory replay in the web app and raises the default max iterations to 250.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.30.1 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.30.1
Adds trajectory replay support in the web app, enabled via the FEATURE_TRAJECTORY_REPLAY feature flag. Not checked yet Tracks accumulated token usage across a session instead of per-request token usage, giving a running total. Not checked yet Raises the default max iterations from the previous limit to 250, allowing agents to tackle longer tasks out of the box. Not checked yet Improves the startup experience for the VSCode extension integration. Not checked yet
4 details in Everything
OpenHands 0.30.0 adds live API cost/token display, web screenshots in headless trajectories, and auto-generated conversation titles.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.30.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.30.0
Displays API cost and token usage live in the frontend during agent sessions. Not checked yet Adds a headless-mode config option to include web screenshots in trajectories for richer audit trails. Not checked yet Auto-generates conversation titles from the first user message. Not checked yet Improved UI for interactive browser action view. Not checked yet Updated file explorer styles for a cleaner workspace experience. Not checked yet
5 details in Everything
└──▷ BREAKING ON UPGRADE
! Download workspace and download files buttons are removed; users are redirected to VS Code instead.! File upload functionality is removed from the UI; users are redirected to VS Code instead.! The continue button is removed from the UI.! Bash SOFT timeout is reduced from 30 to 10 seconds, which may cause previously passing long-running bash commands to time out.
OpenHands 0.29.1 adds a dedicated security microagent for security-focused autonomous tasks.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.29.1 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.29.1
OpenHands 0.29.0 adds Swift Linux microagent support and exposes microagent recall observations to the UI 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. .
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.29.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.29.0
Adds a Swift Linux Installation Microagent for automated Swift environment setup on Linux. Not checked yet Makes microagents available in the event stream as recall observations, accessible from both the UI and the LLM. Not checked yet Adds TestGenEval benchmark support for evaluating test generation capabilities. Not checked yet
3 details in Everything
OpenHands 0.28 adds Docker-in-runtime, Kubernetes cluster support, a CodeAct thinking tool, microagent templates, and JSON structured logging.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.28.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.28.0
└──▷ TRY IT
Enable structured JSON logs so your SIEM SIEM Security information and event management, collecting logs from many systems and correlating events to find suspicious activity. Analysts use it to investigate alerts and incidents. or log aggregator can parse OpenHands output without custom parsers.
copy
$ LOG_JSON=true docker run --rm -e LOG_JSON=true ghcr.io/all-hands-ai/openhands:0.28.0
Adds sound and browser notifications for agent state changes, so practitioners are alerted when long-running tasks complete. Not checked yet New microagent template system lets users create properly structured domain-specific agents. Not checked yet Adds a thinking tool to the CodeAct agent, enabling explicit reasoning steps during task execution. Not checked yet Enables OpenHands to run Docker inside its own runtime, unlocking containerized workflows within the agent sandbox. Not checked yet Enables OpenHands to start a Kubernetes cluster from within its runtime, supporting cluster-level automation tasks. Not checked yet
+1 more
show less
Adds structured JSON logging via the LOG_JSON environment variable for improved observability and log ingestion. Not checked yet
6 details in Everything
└──▷ BREAKING ON UPGRADE
! The default Claude model is now claude-3-7-sonnet-20250219 (previously an older version), which may affect costs, behavior, or rate-limit quotas for deployments that rely on the default.
OpenHands 0.27 adds custom runtime support, an extended config section, and Claude 3.7 model selection.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.27.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.27.0
└──▷ USE IT
Inject arbitrary key-value config into an OpenHands deployment without waiting for first-class support for those keys.
toml
copy
[extended]
my_custom_flag = "value"
another_setting = 42
Switch the agent to Claude 3.7 when you need its extended thinking extended thinking Intermediate reasoning a model writes out before its final answer. Reasoning models generate it automatically and it is often billed as extra output tokens. or latest capabilities.
📍 In the console, go to Settings › Model and select the Anthropic provider, then choose Claude 3.7 from the model list.
Adds an [extended] section in config.toml for arbitrary user-defined configuration keys set on the fly. Not checked yet Supports selecting Claude 3.7 via the Anthropic provider in model settings. Not checked yet Enables advanced users to define and use their own custom runtimes. Not checked yet
3 details in Everything
OpenHands 0.26.0 adds Daytona sandbox runtime, CLI task automation, and UI-configurable memory condensation.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.26.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.26.0
└──▷ HOW TO FIND IT
Enable memory condensation to keep long sessions within 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. context limits without manual intervention.
📍 In the console, go to Settings and enable memory condensation.
Adds Daytona runtime support for managing sandboxes as an alternative execution environment. Not checked yet Enables memory condensation configuration directly from the settings page. Not checked yet Supports running the CLI with a task argument for fully automated, non-interactive execution. Not checked yet Truncates conversation history by half (instead of failing) when the LLM context limit is reached, enabling longer sessions. Not checked yet
4 details in Everything
OpenHands 0.25.0 adds GitLab resolver support, multi-conversation navigation, and a redesigned Settings screen.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.25.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.25.0
Adds GitLab support for the OpenHands resolver, enabling automated issue and PR resolution on GitLab repositories. Not checked yet New multi-conversation navigation lets users switch between different conversations without losing context. Not checked yet Redesigned Settings screen provides an updated UI for managing configuration. Not checked yet Improves filtering of sensitive data from logs, reducing credential and secret exposure in log output. Not checked yet
4 details in Everything
OpenHands 0.24.0 adds dockerless runtime support and success/failure indicators for file operations.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.24.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.24.0
Adds preliminary/experimental support for running OpenHands without Docker, broadening deployment flexibility. Not checked yet Adds success/failure indicators for file read and edit operations, giving agents clearer feedback on action outcomes. Not checked yet Adds comprehensive error logging to PostHog for improved observability of agent failures. Not checked yet
3 details in Everything
OpenHands 0.23.0 adds o3-mini support and live 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. retry visibility with rate-limit resume.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.23.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.23.0
Supports OpenAI o3-mini as a selectable LLM. Not checked yet Shows LLM retries in real time and enables resuming sessions from a Rate-Limited state. Not checked yet
2 details in Everything
OpenHands 0.22.0 adds VisualWebArena evaluation support and drops the Redis dependency.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.22.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.22.0
Adds VisualWebArena evaluation support, enabling visual web-navigation benchmark testing within OpenHands. Not checked yet Removes the Redis dependency, simplifying self-hosted deployments with fewer infrastructure requirements. Not checked yet
2 details in Everything
OpenHands 0.21 adds trajectory replay, downloadable trajectories, multi-language UI, and a dedicated draft-editor 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. config.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.21.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.21.0
└──▷ HOW TO FIND IT
Route edit-draft generation to a cheaper or faster model while keeping your primary LLM for reasoning — set the draft_editor custom LLM config and enable the LLM editor in agent options.
📍 In Settings, add a custom LLM configuration named draft_editor pointing to your preferred model, then enable codeact_enable_llm_editor in the agent options.
Cap compute cost on o1-family models by setting reasoning_effort to a lower tier for less complex tasks.
📍 In Settings › LLM, select an o1-family model and set reasoning_effort to low (or medium/high) to trade off reasoning depth against token cost.
Adds multi-language UI support, broadening accessibility for non-English practitioners. Not checked yet Supports the reasoning_effort parameter for OpenAI o1-family models, giving control over model reasoning depth. Not checked yet Introduces a draft_editor named custom LLM config so a separate, dedicated model can handle edit drafts independently of the primary LLM. Not checked yet Adds trajectory replay in headless mode, enabling automated re-execution of recorded agent sessions. Not checked yet Adds a chat-panel button to download the current agent trajectory for offline review or auditing. Not checked yet
+1 more
show less
Agent now explicitly receives the cloned repo name and path when connecting to a GitHub repository, improving context awareness. Not checked yet
6 details in Everything
└──▷ BREAKING ON UPGRADE
! The trajectories_path config key is renamed to save_trajectory_path; existing configurations using trajectories_path will break on upgrade.
OpenHands 0.20.0 adds a live App tab for interacting with web apps and a conversation info display.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.20.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.20.0
└──▷ HOW TO FIND IT
Preview and interact with a web app your agent is building without leaving the OpenHands UI.
📍 In the OpenHands UI, open your conversation and click the App tab (bottom panel) to view and interact with the running web app in real time.
Adds a new App tab Beta that lets users interact directly with the web app running inside OpenHands. Not checked yet Displays current conversation info in the bottom-right corner of the UI. Not checked yet
2 details in Everything
OpenHands 0.19.0 adds custom microagents in headless/CLI modes, toggleable 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. , multi-conversation support, and runtime size configuration.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.19.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.19.0
Supports custom microagents in headless and CLI modes, extending agent customization beyond the UI. Not checked yet Enables or disables function calling via user configuration, giving practitioners control over 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-use behavior. Not checked yet Adds rate-limit error visibility in the UI and agent state, surfacing provider throttling in real time. Not checked yet Introduces multi-conversation feature (behind a feature flag), allowing parallel conversation management. Not checked yet Adds runtime size configuration, letting users tune the sandbox compute footprint per session. Not checked yet
5 details in Everything
└──▷ BREAKING ON UPGRADE
! Developer mode now defaults to storing data on the filesystem instead of in-memory, which changes persistence behavior for existing developer-mode deployments.
OpenHands 0.18.0 adds per-repo prompt customization, resizable panels, and server-side settings storage.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.18.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.18.0
└──▷ HOW TO FIND IT
Customize the agent's system prompt for a specific repository so it follows project-specific conventions on every run.
📍 In your repository root, create a .openhands directory and add your prompt customization files there. OpenHands will automatically detect and apply them when operating on that repo.
Adds .openhands directory support for customizing agent prompts on a per-repository basis. Not checked yet Introduces a resizable and collapsible panel layout in the UI for a more flexible workspace. Not checked yet Enables workspace downloads on mobile and browsers lacking the Directory API. Not checked yet Migrates settings storage from browser localStorage to the server API, enabling persistent settings across devices and browsers. Not checked yet Adds navbar tooltips for improved UI discoverability. Not checked yet
5 details in Everything
└──▷ BREAKING ON UPGRADE
! File editing functionality has been removed from the OpenHands UI — workflows that relied on in-UI file editing must use an alternative method.
OpenHands 0.17 adds custom PR titles, per-conversation URLs, and a homepage recent-conversation view.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.17.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.17.0
Supports custom pull request titles when using the GitHub issue resolver. Not checked yet Each conversation now gets its own unique URL, enabling direct linking and bookmarking. Not checked yet Shows the most recent conversation on the homepage for quick resumption. Not checked yet Adds TheAgentCompany evaluation harness for benchmarking agent performance. Not checked yet Settings store type is now defined in openhands_config instead of the main config. Not checked yet
5 details in Everything
└──▷ BREAKING ON UPGRADE
! Settings store type is now defined in openhands_config rather than the main config — configurations relying on the main config for this setting will need to be updated.
OpenHands 0.16.1 adds file-by-file download with progress and backend config saving infrastructure.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.16.1 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.16.1
Adds file-by-file download with progress tracking for workspace files. Not checked yet Introduces infrastructure for saving configuration on the backend rather than the frontend. Not checked yet
2 details in Everything
OpenHands 0.16.0 updates GitHub Authentication flow.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.16.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.16.0
OpenHands 0.15.3 raises default max_iterations to 500 and resets turn budget per user interaction
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.15.3 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.15.3
Adds color coding to the Ran a Command UI indicator based on command return value, making pass/fail status immediately visible. Not checked yet Increases default max_iterations to 500, enabling agents to tackle longer, more complex tasks before hitting the iteration limit. Not checked yet Measures the turn limit from the last user interaction rather than session start, giving agents a full budget after each user message. Not checked yet Persists jwt_secret in the config file so the secret survives service restarts without manual reconfiguration. Not checked yet
4 details in Everything
OpenHands 0.15.2 adds workspace-level custom MicroAgents, resolver code-review requests, and browser observations in chat.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.15.2 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.15.2
Adds support for Custom MicroAgents placed directly in the connected workspace, enabling project-specific agent behavior without global configuration. Not checked yet Enables the resolver to automatically request a code review from the person who initiated it after completing a PR. Not checked yet Surfaces browser observations (what the agent sees while web browsing) directly in the chat interface for real-time visibility. Not checked yet
3 details in Everything
OpenHands 0.15.1 adds command-run notifications and multiline message support in headless mode.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.15.1 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.15.1
Shows a status message in the UI each time the agent executes a command, giving real-time visibility into agent actions. Not checked yet Supports multiline user messages, and provides a default user message when running in headless mode. Not checked yet
2 details in Everything
OpenHands 0.15.0 adds SocketIO comms, BrowseURLAction in CodeAct, custom resolver sandboxes, and multi-tab session sharing.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.15.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.15.0
Switches underlying client-server communication to SocketIO for more robust real-time messaging. Not checked yet Adds BrowseURLAction to the CodeAct agent, enabling it to navigate URLs as part of agentic workflows. Not checked yet Adds an alignment checker for the browsing agent to verify page-interaction accuracy. Not checked yet Supports custom sandbox images for the GitHub issue resolver. Not checked yet Adds a microagent to automatically handle npm timeout issues during code execution. Not checked yet
+2 more
show less
Allows multiple browser tabs to connect to and share the same agent session simultaneously. Not checked yet Runtimes are no longer kept alive by default, reducing idle resource consumption. Not checked yet
7 details in Everything
└──▷ BREAKING ON UPGRADE
! Runtimes are no longer kept alive by default — existing deployments that relied on persistent runtimes between sessions will need to reconfigure this behavior explicitly.
OpenHands 0.14.3 adds a Push & Create PR button and the commit0 benchmark.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.14.3 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.14.3
└──▷ HOW TO FIND IT
After an agent completes its work, push the branch and open a PR without leaving the UI.
📍 In the OpenHands UI, after a session completes, click the Push & Create PR button to push the current branch and create a pull request in your connected repository.
New Push & Create PR UI button lets users push changes and open a pull request directly from the OpenHands interface. Not checked yet Adds the commit0 benchmark for evaluating agent performance. Not checked yet Server process now runs at highest system priority, improving responsiveness under load. Not checked yet
3 details in Everything
OpenHands 0.14.1 adds a push-to-GitHub button directly in the chat interface.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.14.1 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.14.1
└──▷ HOW TO FIND IT
After completing a coding task in a session, push your changes to GitHub without leaving the chat interface.
📍 In the chat interface, click the Push to GitHub button to push your current changes to the configured GitHub repository.
OpenHands 0.14.0 adds an embedded VSCode editor, GitHub issue resolver integration, and universal 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. shim for non-native LLMs LLMs 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. .
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.14.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.14.0
└──▷ HOW TO FIND IT
Open the embedded VSCode editor from within the OpenHands UI to inspect and edit agent-modified files without leaving the interface.
📍 In the OpenHands UI, open an active session and click the VSCode panel to launch the integrated editor against the running runtime.
Adds a full VSCode editor directly inside the OpenHands runtime and UI, giving agents and users an integrated code editing experience. Not checked yet Integrates the OpenHands resolver into the main repository, enabling automated GitHub issue resolution workflows natively. Not checked yet Adds a utility that automatically converts function-calling 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 to non-function-calling format, broadening compatible model support. Not checked yet
3 details in Everything
OpenHands 0.13.1 adds workspace media previews and improved GitHub push support.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.13.1 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.13.1
Supports displaying images, videos, and PDFs directly in the workspace UI. Not checked yet Improves pushing changes to GitHub from within a session. Not checked yet
2 details in Everything
OpenHands 0.13.0 adds drag-and-drop image input to chat and a loading indicator for ZIP downloads.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.13.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.13.0
Supports dragging images directly into the chat input for multimodal multimodal Able to work with more than one type of input or output, such as text, images and audio. A multimodal model can combine information across those formats. interactions. Not checked yet Shows a loading indicator when downloading a workspace ZIP file. Not checked yet New empty-state design for the chat interface. Not checked yet
3 details in Everything
OpenHands 0.12.2 gives CodeAct native web browsing, adds PostHog analytics, and integrates DiscoveryBench.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.12.2 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.12.2
Adds PostHog analytics integration for tracking usage and agent behavior. Not checked yet Enables CodeAct agent to perform web browsing natively — no longer delegates to a separate BrowsingAgent. Not checked yet Adds DiscoveryBench integration for benchmarking OpenHands agents on data-discovery tasks. Not checked yet
3 details in Everything
OpenHands 0.12.0 adds 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. to CodeAct 2.1 and prompt caching for Claude 3.5 Sonnet, hitting 53% on SWE-Bench Verified.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.12.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.12.0
Adds function calling support to the CodeAct 2.1 agent, boosting benchmark performance to 53% on SWE-Bench Verified and 41.7% on SWE-Bench Lite — the highest score among open-source AI software engineering systems. Not checked yet Supports prompt caching for claude-3-5-sonnet-20241022, reducing latency and cost on repeated context. Not checked yet Surfaces backend error messages directly in the chat UI so practitioners see failure details without digging through logs. Not checked yet
3 details in Everything
OpenHands 0.11.0 improves 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. -powered agent file editing.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.11.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.11.0
OpenHands 0.10.0 ships a redesigned UI and adds Modal as a new sandbox runtime.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.10.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.10.0
New OpenHands UI with a redesigned interface for interacting with agents. Not checked yet Adds Modal sandbox support as an alternative runtime backend for agent execution. Not checked yet
2 details in Everything
OpenHands 0.9.8 lets you specify a custom config.toml path via command-line arguments.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.9.8 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.9.8
Supports specifying the config.toml location via command-line arguments, enabling flexible multi-environment deployments. Not checked yet
1 detail in Everything
OpenHands 0.9.5 adds more granular startup status messages in the UI.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.9.5 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.9.5
OpenHands 0.9.4 adds a --version flag to the CLI and reduces Gemini safety thresholds.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.9.4 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.9.4
Adds --version flag to the OpenHands CLI for quick version inspection. Not checked yet Reduces safety thresholds for Gemini models, enabling broader task execution with Gemini backends. Not checked yet
2 details in Everything
OpenHands 0.9.3 adds interactive command support and a redesigned settings modal.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.9.3 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.9.3
Supports interactive terminal commands, enabling agents to handle prompts that require user input mid-execution. Not checked yet Redesigned settings modal with an updated UI for configuring the service. Not checked yet Makes prompt caching opt-in rather than enabled by default, giving users explicit control over caching behavior. Not checked yet
3 details in Everything
└──▷ BREAKING ON UPGRADE
! Prompt caching is no longer enabled by default; users who relied on it must explicitly opt in after upgrading.
OpenHands 0.9.1 adds a CLI mode, prompt caching for Sonnet/Haiku, and specialized micro agents for CodeAct.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.9.1 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.9.1
Adds CLI mode, enabling OpenHands to be driven from the command line without the web UI. Not checked yet Adds prompt caching to CodeActAgent for Anthropic Sonnet and Haiku models, reducing latency and token costs on repeated context. Not checked yet Introduces micro agents as specializations of CodeAct, allowing targeted agent behaviors for specific tasks. Not checked yet Improves model selection UI in settings for easier 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. configuration. Not checked yet
4 details in Everything
OpenHands 0.9.0 adds vision 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. support, a Security Analyzer, async LLM calls, and switches to EventStream Runtime by default.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.9.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.9.0
Adds vision input support for LLMs LLMs 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. with vision capabilities, enabling image-aware agent interactions. Not checked yet Adds a Security Analyzer functionality for automated security review workflows. Not checked yet Enhances the LLM class with async support for non-blocking model calls. Not checked yet Implements copy_to for Runtime, allowing files to be pushed directly into the agent's runtime environment. Not checked yet Supports injecting additional dependencies into the OpenHands runtime Docker image at launch. Not checked yet
+7 more
show less
Switches the default runtime to EventStream Runtime. Not checked yet Clears conversation history at the start of each new task to avoid context bleed between sessions. Not checked yet Adds collapsible resizers in the UI for a more flexible workspace layout. Not checked yet Integrates the Aider Benchmark for agent evaluation. Not checked yet Agents can now accept per-agent configuration via agent config, enabling fine-grained control. Not checked yet Adds warning messages when a config file contains unknown keys, helping catch misconfiguration early. Not checked yet Adds error observations to the CodeAct SWE agent so it can recover from execution failures. Not checked yet
12 details in Everything
└──▷ BREAKING ON UPGRADE
! The project is renamed from OpenDevin to OpenHands — any references to the old name in configs, Docker images, or scripts will need updating.! The default runtime is now EventStream Runtime instead of the previous default — existing setups relying on the old default runtime behavior may behave differently without explicit configuration.
OpenHands 0.8.3 adds a UI toggle for agent selection and Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. support for GitHub Codespaces.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.8.3 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.8.3
Adds a UI toggle to enable or disable agent selection (defaults to off, hiding agent selection from end users). Not checked yet Adds Dev Container configuration for GitHub Codespaces, enabling one-click cloud development environments. Not checked yet
2 details in Everything
OpenHands 0.8.2 adds headless controller mode and SWE-Bench custom Docker namespace support for eval workflows.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.8.2 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.8.2
Adds headless_mode for the Controller, enabling programmatic agent runs without a UI session. Not checked yet Supports pulling SWE-Bench evaluation environments from a custom Docker namespace. Not checked yet Shrinks the runtime image size, reducing pull times and storage overhead. Not checked yet
3 details in Everything
└──▷ BREAKING ON UPGRADE
! The Monologue agent has been removed and is no longer available as an agent type.! Global config has been removed from the sandbox, agent controller, and related components — configurations previously set globally must now be passed explicitly.
OpenHands 0.8.0 adds user confirmation mode, per-agent 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. config, EventStream runtime with browser and Jupyter support, and restores append_file.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.8.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.8.0
Adds append_file action back to the agent's file-editing toolkit, enabling agents to append content to files without overwriting them. Not checked yet Implements user confirmation mode: when enabled, OpenHands requests explicit user confirmation before executing bash or Python code. Not checked yet Enables per-agent LLM configuration, allowing each delegated agent to use a distinct LLM config set from the UI. Not checked yet Introduces EventStreamRuntime with an agnostic sandbox, adding Jupyter notebook support and browser interaction to the runtime client. Not checked yet Adds a 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. runtime (od-client-runtime) as a new runtime backend option. Not checked yet
+3 more
show less
Displays real-time build logs for the agnostic sandbox image during construction. Not checked yet Deprecates the recall action and search_memory capability, removing them from the agent action surface. Not checked yet Removes support for Background Commands from the runtime architecture. Not checked yet
8 details in Everything
└──▷ BREAKING ON UPGRADE
! Background Commands are no longer supported — any working setup relying on background command execution will break on upgrade.! The recall action and search_memory are deprecated and removed — agents or configurations invoking these will break on upgrade.
OpenHands 0.7.1 adds local Ollama model support, CLI session restore, config templates, and DEBUG logging control.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.7.1 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.7.1
Adds make run-wsl target for running OpenHands under Windows Subsystem for Linux. Not checked yet Enables DEBUG-level logging via a config setting, giving operators fine-grained log control without code changes. Not checked yet Adds a config template feature to bootstrap new configuration files. Not checked yet Adds support for local Ollama models, letting teams run inference entirely on-premises without an external 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. Not checked yet Adds optional CLI session restore, allowing interrupted CLI sessions to be resumed. Not checked yet
+7 more
show less
Enables concurrent CLI operation, improving throughput for command-line workflows. Not checked yet Removes the MAX_CHARS traffic-control limit, lifting a cap on observation size. Not checked yet Allows MAX_ITERATIONS and MAX_BUDGET_PER_TASK to be overridden directly from the web GUI GUI A graphical user interface, controls such as windows, buttons and menus. It lets people use software visually rather than entering commands. . Not checked yet Improves code editing in the CodeAct agent by using diff-based patching for more reliable file edits. Not checked yet Removes the Docker exec sandbox backend, consolidating on the SSH-based sandbox. Not checked yet Adds automatic LLM retry on internal server errors, improving resilience against transient upstream failures. Not checked yet Docker builds for stable releases now also publish a latest tag, simplifying image pulls. Not checked yet
12 details in Everything
└──▷ BREAKING ON UPGRADE
! The Docker exec sandbox box has been removed; deployments relying on it must migrate to the SSH-based sandbox.! The MAX_CHARS traffic-control setting has been removed; any configuration referencing it will no longer have effect.! Global CLI arguments have been removed; workflows depending on those global args must be updated.
OpenHands 0.7.0 adds WebArena, MiniWoB++, GPQA, ToolQA, Gorilla APIBench, and BioCoder benchmarks plus CodeActAgent browser delegation.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.7.0 https://github.com/All-Hands-AI/OpenHands.git
$ git checkout 0.7.0
Renames the OPENDEVIN_WORKSPACE environment variable to WORKSPACE_BASE across the codebase. Not checked yet Adds append_file skill to agentskills, enabling agents to append content to files. Not checked yet Adds WebArena and MiniWoB++ benchmark support with related architecture changes for browser-based evaluation. Not checked yet Adds GPQA benchmark evaluation support. Not checked yet Adds ToolQA benchmark support. Not checked yet
+8 more
show less
Adds Gorilla APIBench benchmark support. Not checked yet Adds BioCoder benchmark integration. Not checked yet Adds SWE-Bench Docker-based evaluation (SWEBench-docker eval ). Not checked yet Enables CodeActAgent to delegate browser tasks to BrowsingAgent, allowing the code agent to hand off web interactions to a specialized sub-agent. Not checked yet Revamps AgentRejectAction and enables ManagerAgent to handle task rejections gracefully. Not checked yet Adds evaluation time-travel capability, allowing benchmark evaluation to target a specific historical version of the codebase. Not checked yet Makes plugin sandbox Dockerfile sandbox-agnostic, broadening deployment flexibility. Not checked yet Adjusts the stuck-step detection threshold to 3 consecutive identical steps before halting. Not checked yet
13 details in Everything
└──▷ BREAKING ON UPGRADE
! The OPENDEVIN_WORKSPACE environment variable is renamed to WORKSPACE_BASE; existing configurations referencing OPENDEVIN_WORKSPACE will no longer be recognized.! The github-token config key is removed; configurations using it will no longer work.