HiClaw

HiClaw is an open-source platform for running teams of AI agents that coordinate with each other and with human users, all inside a chat room interface. Instead of running AI agents in the background where you cannot see what they are doing, HiClaw places you in the same room as the agents so you can watch every message, step in at any point, or redirect them when something goes wrong.

The architecture separates agents into a Manager and multiple Workers. The Manager handles overall task planning and delegates subtasks to Workers. Workers are specialized agents that can write and run code, answer questions, or take other actions depending on which runtime they use. HiClaw supports multiple agent runtimes in the same room, including options called OpenClaw, QwenPaw, and Hermes, each suited to different kinds of tasks.

The chat layer runs on Matrix, an open messaging protocol that you can host yourself. The Element web client is included and connects to a built-in Matrix server, so there is no dependency on Slack, DingTalk, or any external messaging service. All communication between agents and humans happens through this interface, making the collaboration visible and auditable.

On the security side, real API keys and credentials stay inside an AI gateway component. The Worker agents only receive limited-scope consumer tokens, so they cannot access or leak the underlying credentials even if they are given malicious instructions.

Installation is a single command that runs an installer script on macOS, Linux, or Windows. The script sets up the AI gateway, Matrix server, file storage, web client, and Manager Agent automatically. A Kubernetes Helm chart is available for shared or production deployments. Minimum requirements are 2 CPU cores and 4 GB of RAM, with 4 cores and 8 GB recommended when running multiple Workers. The project works with any OpenAI-compatible LLM API.