unohee/OpenSwarm

What it solves

OpenSwarm is an autonomous AI agent orchestrator designed to act as a team of code workers. It automates the process of picking up software issues from trackers like Linear or a local store, implementing fixes, and verifying them through a structured pipeline of worker and reviewer agents, reducing the manual effort required for routine bug fixes and feature implementation.

How it works

OpenSwarm uses a "PairPipeline" architecture where a Worker agent implements a task and a Reviewer agent validates the result. This process is supported by several components:

• Multi-Provider Adapters: It can use various LLMs via OpenAI Codex/GPT, OpenRouter, local models (Ollama, LM Studio), or the Claude Code CLI.
• Task Integration: It integrates with Linear or a local SQLite issue store to manage tasks.
• Deterministic Verification: It runs repository-specific test and typecheck commands to ensure changes don't introduce regressions.
• Memory & Knowledge: It uses LanceDB for long-term memory and a SQLite-backed Code Registry to track functions, classes, and types across multiple languages.
• Notification System: It reports progress via Discord, Slack, Telegram, or webhooks.

Who it’s for

It is built for developers and engineering teams who want to automate software maintenance, bug fixing, and code auditing using a swarm of autonomous AI agents.

Highlights

• Hybrid Model Routing: Supports combining frontier models for diagnosis and lightweight models for implementation to optimize cost.
• Extensible Provider System: Seamlessly switch between cloud APIs and local LLMs at runtime.
• Code Registry & BS Detector: Scans repositories to map entities and identify "bad code smells" (BS patterns).
• Autonomous Daemon: Can run as a background service that monitors issue trackers and automatically submits PRs.
• TUI Interface: Includes a terminal user interface for chatting and managing projects and tasks.