A developer has built OpenSwarm, a CLI orchestrator that coordinates multiple Claude Code instances as autonomous agents to handle real Linear issues. The system creates a Worker/Reviewer/Test/Documenter pipeline that pulls tasks from Linear, processes them through Claude CLI instances, and reports progress via Discord. Built on Node.js 22, it uses LanceDB with multilingual-e5 embeddings for long-term memory and builds a simple code knowledge graph for impact analysis.

Currently powering the creator's solo development workflow across trading infrastructure and LLM tools, OpenSwarm is still early with rough edges around safety, scaling, and task decomposition. The orchestrator can auto-iterate on existing PRs, monitor long-running jobs, and expose everything through a Discord bot interface for status, dispatch, scheduling, and logs. Configuration supports environment variables and Zod validation, with agents defined by role models and heartbeat intervals.

The project seeks feedback on missing features for team adoption, potential failure modes in autonomous code agents, and ideas for better memory/knowledge graph integration in real-world repositories. With 12 stars on GitHub, the open-source tool represents an ambitious attempt to create a practical "AI dev team" that integrates directly into existing developer workflows rather than running isolated toy tasks.