Awesome AI Agents › Multi-Agent Frameworks

AlexAnys/opencrew

⭐ 494 Shell repository created 2026-02-08

OpenCrew is a configuration and protocol layer built on OpenClaw that splits a single overloaded assistant into a team of domain-specific agents, each owning its own area and its own context. The organising idea maps a chat workspace onto a company: a channel or group is a role, a thread is a task, and the visible channel list becomes the task overview that a single long-running session cannot provide. Slack, Feishu and Discord are all supported as the command centre. The project addresses four failure modes the author observed in single-agent use: context bloat that makes the assistant dull over a long session, loss of overview when several projects run in parallel, constant confirmation prompts because the agent does not know what it may decide alone, and lessons that stay buried in chat history. The answers are per-domain agents, an explicit autonomy-level mechanism tied to intent alignment, a three-layer knowledge pipeline that promotes conversation into structured summaries and then into reusable knowledge, and a dedicated maintenance agent that audits the others for drift. The second version of the agent-to-agent protocol removes the earlier limitation that all agents shared one bot identity and could therefore only delegate one way. Giving at least one coordinating agent its own application lets it join the channels of executing agents and hold real discussions there, with review and consensus in the channel and the actual work exchanged through Markdown files, before the human reviews the final output. Suggested standalone roles are a chief of staff representing user intent, a planner that expands requirements into acceptance criteria, and a separate quality evaluator, on the argument that an agent grading its own output is too lenient. Documentation is primarily Chinese with an English translation, and the deployment is shell-based.

https://github.com/AlexAnys/opencrew

a2a-protocolagent-deploymentagent-orchestrationai-agentsai-assistantai-productivityai-teamautonomous-agentsclawdbotcontext-managementknowledge-managementllmmulti-agentno-codeopenclawsession-managementslackslack-botteam-managementworkflow-automation