About AgentCrew
AgentCrew is an open source multi agent assistant framework that lets one person run a team of specialised AI agents for coding, research and automation. Each agent gets its own system prompt and tools, and agents can transfer a conversation to one another or delegate tasks in parallel, so a research agent can gather material, an analyst can process it and a writer can report on it.
It is aimed at developers and technical users comfortable with Python 3.12 or later. AgentCrew connects to Claude, GPT, Gemini, GitHub Copilot, OpenAI Codex, DeepInfra, Together AI, OpenCode or any OpenAI compatible endpoint, and different models can be mixed within one project. Built in tools cover browser automation, file editing, code analysis, shell commands, web search, clipboard, voice and MCP servers. It runs as a desktop GUI, a terminal chat, a headless job runner that can enforce JSON schema output for CI pipelines, or an A2A HTTP server. Semantic memory powered by ChromaDB retrieves relevant context and supports forgetting by topic, and an evolve command reviews an agent's memory to propose system prompt improvements for approval. Conversations can be rolled back, forked, consolidated to save tokens and resumed, and agent configurations can be exported as TOML or JSON for sharing.
AgentCrew is free under the Apache 2.0 license and has no paid plan; users pay their chosen model providers directly through their own API keys.
The project is in beta and is backed by Saigon Technology. Safety is on by default: every tool call requires approval, commands are rate limited, paths are validated and actions are logged for audit.