Researchers at HKU’s Data Intelligence Lab have released AutoSwarm, an MIT-licensed Python tool that coordinates several coding agents, shares context among them, and runs subtasks in parallel. The repository had more than 2,500 GitHub stars at the time of writing, and a companion paper describes its coordination algorithms.
AutoSwarm addresses a growing source of friction in AI-assisted development. Tools such as Claude Code, Cursor, Codex, and Copilot maintain separate contexts, expose different capabilities, and connect to different data sources. Developers working across several agents often repeat repository instructions, transfer prompts manually, and combine the resulting code or analysis themselves.
One prompt, many workers
After a developer submits a task through a supported host, AutoSwarm selects agents by capability and coordinates their work. Its workflow covers five stages:
- Break the request into smaller tasks.
- Assign each task to an appropriate installed agent.
- Share relevant context and memory across workers.
- Run independent tasks concurrently when possible.
- Verify and combine the outputs into one response.
The project’s desktop interface illustrates the model with a release task divided among Claude Code, Cursor, OpenCode, Kimi Code, Copilot CLI, and Grok Build. Each subtask has its own status and timestamps, giving the user one place to follow parallel work.