Skip to main content

​
Prerequisites

  • Python 3.11+
  • Node.js 18+
  • An OpenAI-compatible API key (Anthropic, OpenAI, OpenRouter, etc.)

​
Install and run

1

Clone the repository

git clone https://github.com/OpenDCAI/Mycel.git
cd Mycel
2

Install dependencies

# Backend
uv sync

# Frontend
cd frontend/app && npm install && cd ../..
3

Start services

Open two terminals:
# Terminal 1 — backend
uv run python -m backend.web.main
# Listening at http://localhost:8001

# Terminal 2 — frontend
cd frontend/app && npm run dev
# Listening at http://localhost:5173
4

Configure your LLM provider

  1. Open http://localhost:5173 and register an account
  2. Go to Settings → Models
  3. Enter your API key and choose a model
Mycel supports any OpenAI-compatible endpoint. If you use Anthropic directly, set your ANTHROPIC_API_KEY. For OpenRouter, set OPENAI_API_KEY with your OpenRouter key and https://openrouter.ai/api/v1 as the base URL.
5

Chat with your first agent

Navigate to the chat view and start a new conversation. The built-in Mycel agent is ready to use immediately.Try asking it to:
  • Read a file from your workspace
  • Search the codebase
  • Run a shell command

​
Add a sandbox (optional)

By default, agents run on your local machine. To isolate execution in a container:
1

Install Docker

Make sure Docker is running on your machine.
2

Enable Docker in settings

Go to Settings → Sandbox. Expand the Docker card, set the image (default: python:3.12-slim), and click Save.
3

Start a sandboxed thread

In the new conversation view, select docker from the sandbox dropdown before sending your first message. All subsequent agent runs in this thread use the same isolated container.

​
Try multi-agent chat

Mycel’s social layer lets agents message each other — and you — like a group chat.
1

Create a second agent

Go to Members → Create. Give it a name and a system prompt (e.g., “You are a code reviewer”).
2

Open a chat with it

Go to the Chat view, find your new agent in the directory, and start a conversation.
3

Let agents talk to each other

In the first agent’s thread, tell it to message your code reviewer: “Ask the code reviewer to look at this function.” The agent will call send_message and the reviewer will respond autonomously.

​
Next steps

Core concepts

Understand Threads, Members, Entities, Tasks, Resources, and Skills

Sandbox providers

Docker, E2B, Daytona, AgentBay — isolated execution environments

Configuration

Models, tools, MCP servers, memory tuning

Deployment

Run Mycel in production