HKUDS unifies your coding agents with AutoSwarm

AutoSwarm is an open-source multi-agent coordination tool from HKUDS that turns your fragmented AI agents into a single collaborative team.

Right now, no single agent handles everything well. You might write code in Cursor or Claude Code, run terminal commands in DeepSeek Harness, and query another model for research. Each tool traps its own context and memory, forcing you to constantly copy and paste prompts, manually route subtasks, and glue outputs together.

What it does

AutoSwarm eliminates that friction by acting as a central coordinator across your existing tools.

First, it routes tasks based on strengths. AutoSwarm connects directly to a wide roster of coding and CLI tools, including Codex, Claude Code, OpenCode, Cursor, GitHub Copilot CLI, Meta Muse Code, CodeBuddy, Hermes, Nanobot, DeepSeek Harness, Grok Build, and Kimi Code. Instead of treating every agent as a generalist, it delegates subtasks to the tool best suited for the job.

Second, it unifies context and memory across those boundaries. AutoSwarm bridges different workflows and local environments, runs independent subtasks in parallel for speed, sequences dependent steps, and validates the final deliverables before synthesizing them into one coherent answer.

How to get started

AutoSwarm runs across Windows, macOS, and Linux. You can interact with it through desktop applications, CLI terminals, IDE extensions, or any host agent supporting Skills.

To install it, you need a Python 3.10+ environment. The recommended path is cloning the GitHub repository and running uv tool install --force ., followed by autoswarm setup --host codex (or another supported host like claude-code or cursor). Standard pip installation in a dedicated environment is also supported.

Once installed, you simply use any agent as your entry point. Ask it to read 100 documents and summarize conflicting approaches with AutoSwarm, and it automatically verifies installed local collaborators and orchestrates the workload. You can run autoswarm doctor to probe model responses and resume states across providers.

Why it matters

If your daily workflow already involves juggling multiple coding assistants and terminal CLIs, AutoSwarm is worth installing today. Instead of pretending one frontier model can replace your entire toolkit, it takes the practical route by acting as a coordination fabric over the software you already use.

We think this lightweight plug-in architecture is the right way forward for agent orchestration. Keep in mind that AutoSwarm relies heavily on local CLI compatibility and native authentication, so expect to spend a few minutes verifying token permissions and environment paths during initial setup.