langchain-ai/open-swe

Features

• Automated Software Engineering Agents - Coordinates asynchronous AI agents to automate end-to-end software development tasks from analysis to submission.
• AI Agent Orchestration Frameworks - Provides a runtime framework for managing autonomous agent loops, intercepting model calls, and coordinating tool discovery.
• Stateful Agent Orchestration - Implements a stateful orchestration framework to coordinate complex, long-running AI software engineering workflows.
• Hybrid Short-and-Long Term Memory - Implements memory architectures that integrate active session context with persistent long-term storage for agent reasoning.
• Agent State Persistence - Saves and restores the execution state of agents to a database to ensure continuity across interruptions.
• AI Agent Orchestrators - Coordinates groups of specialized sub-agents using structured workflows to complete complex software engineering projects.
• Tool Integrations - Provides integration with version control and issue tracking tools via webhooks to automate engineering tasks.
• Human-in-the-Loop Tools - Provides tools for users to interact with and guide agents by modifying their internal state during active execution.
• GitHub Event Integrations - Starts automated coding tasks triggered by tags in GitHub issue titles, bodies, comments, or reviews.
• Linear Event Integrations - Launches coding tasks by listening for mentions within Linear project comments.
• Event-Driven Task Triggers - Launches automation tasks through mentions in chat applications and comments on issues or pull requests.
• Omnichannel Automation Triggers - Implements a webhook-driven interface that triggers engineering workflows via mentions in GitHub, Slack, and Linear.
• Repository File Management Tools - Provides utilities for reading, writing, editing, and searching files within a target code repository.
• Webhook Triggers - Exposes HTTP webhook endpoints that allow external platforms to trigger automated project agent workflows.
• Webhook Triggers - Implements webhook endpoints to trigger agent workflows based on events from GitHub, Slack, and Linear.
• Code Execution Sandboxes - Provisions secure, isolated environments for agents to execute shell commands and edit code without risking the host system.
• Remote Command Execution - Enables the execution of shell commands within remote Linux environments to provide full system access to the agent.
• Workflow Sync Integrations - Syncs agent activity and triggers workflows based on mentions and comments in external communication tools.
• Slack Integrations - Initiates engineering tasks through Slack app mentions and direct messages using custom routing.
• Git Access Management - Manages authenticated git traffic using a combination of user tokens and application installation proxies.
• Isolated Execution Sandboxes - Spawns resource-constrained cloud environments for each agent thread, managing lifecycles and health checks.
• Credential Injection Proxies - Injects authentication tokens into sandbox traffic via a proxy to interact with repositories without storing raw tokens.
• Remote Sandbox Isolation - Provides isolated remote Linux environments for secure execution of shell commands and file modifications.
• External Tool Integrations - Integrates with external team communication and project tracking tools to trigger and sync automated workflows.
• Code Style Guides - Loads custom coding standards and review prompts at runtime to ensure consistent feedback across repositories.
• Coding Convention Specifications - Enforces architectural decisions and coding standards by reading root markdown files into agent prompts.
• Parallel Subagent Orchestrators - Coordinates multiple independent subagents to handle discrete subtasks simultaneously using separate operational states.
• Project Context Managers - Injects repository conventions and full thread history to provide the agent with necessary project context.
• Automated Code Review - Automatically analyzes code changes to identify issues and enforce repository-specific style guidelines.
• Agent Tool Definitions - Expands agent capabilities by defining custom functional tools and assigning them based on trigger source.
• Execution Message Injection - Allows users to inject new instructions into active agent threads mid-execution to steer the model's processing.
• Execution Hooks - Injects custom logic into the agent loop to intercept messages and handle tool errors using middleware hooks.
• Agent Execution Tracing - Provides tools for inspecting the history of thoughts, tool calls, and state transitions during agentic workflows.
• Agent Configuration Management - Manages authentication, user profiles, and model configurations through a centralized administrative system.
• Style Learning - Analyzes historical pull request data to automatically generate and refine style prompts for the reviewer agent.
• System Prompt Configurations - Provides modular system prompt configurations to define roles and behavioral parameters for the coding agents.
• Pre-Commit Validations - Runs linters, formatters, and tests within a sandbox to ensure changes are correct before committing them.
• Agent Workflow Interception - Runs deterministic middleware hooks to inject messages, handle tool errors, or notify users of step limits.
• Sandbox Provider Interfaces - Provides a standard protocol to extend the isolation layer with new remote sandbox backend providers.
• Sandbox Configuration - Defines isolated runtimes by specifying provider types, resource limits, and pre-configured snapshots.
• Repository Resolution Strategies - Implements priority-based lookup of tokens and metadata to identify the correct target repository for agent tasks.
• Stateful Agent Hosting - Provides infrastructure dedicated to hosting long-running AI agents while maintaining state across operations.
• Incremental Streaming - Streams partial task outputs to the client in real-time, allowing users to monitor agent progress incrementally.
• Proxy-Based Credential Injection - Secures git operations by injecting authentication tokens into sandbox network traffic via a proxy.
• Hierarchical Subagent Orchestrations - Manages multi-level agent delegation by spawning independent child agents to process subtasks in parallel.
• Custom Middleware Implementations - Injects custom deterministic logic into the agent loop via hooks that run before or after model execution.
• Read-Only Analysis Modes - Performs non-destructive inspections of pull requests to identify potential issues without altering the source code.
• Administrative Management Dashboards - Provides a dedicated web interface for managing team defaults, repository access, and agent settings.
• Code Quality Tools - Runs linting and automated formatting on source files to ensure consistent style and identify potential bugs.
• Agentic Coding Tools - Asynchronous coding agent for software engineering.
• Application Examples - Supports autonomous, long-running software engineering tasks.
• Coding Agents and IDEs - Async cloud-hosted framework for coding agents.

Star history