mycel

AI agents are powerful alone — chaotic when they work together. mycel fixes that.

Coordinate teams of Claude, Gemini, Cursor, and other AI agents with isolated worktrees, shared channels, and cost controls. One binary. No login. MIT licensed.

Install

# macOS / Linux
curl -fsSL https://raw.githubusercontent.com/rpuneet/mycel/main/scripts/install.sh | bash

# Homebrew
brew install rpuneet/mycel/mycel

# Go
go install github.com/rpuneet/mycel/cmd/mycel@latest

# From source
git clone https://github.com/rpuneet/mycel && cd mycel && make install-local-mycel

Prerequisites: Go 1.25+, tmux, git. For the web UI / desktop app: Bun.

Quick Start

mycel up                      # Start server + web UI on localhost:9374 (bootstraps ~/.mycel)
mycel agent create eng-01 \
  --role engineer \
  --tool claude               # Spawn an agent
mycel status                  # See what's running
mycel agent peek eng-01       # Watch agent output
mycel cost show               # Check spending

Open http://localhost:9374 for the web dashboard.

What mycel Does

Without mycel: One agent at a time. Context lost between sessions. Merge conflicts from parallel edits. No visibility. Surprise bills.

With mycel: Multiple agents in parallel. Each gets its own git worktree — zero conflicts. Persistent channels for structured communication. Per-agent budgets with hard stops. Everything observable in real time.

How It Works

1. Isolated by design — Each agent runs in its own tmux session (or Docker container) with a dedicated git worktree. They can't step on each other's work.

2. Structured communication — Agents talk through persistent channels with mentions, reviews, and handoffs. Not through you.

3. Full visibility — Costs, activity, resource usage — all in real time through the CLI, web dashboard, or desktop app.

Supported Agents

Agent	Status
Claude Code	Fully supported
Gemini CLI	Fully supported
Cursor	Supported
Codex	Supported
Custom	Any CLI tool via provider config

Commands

Core

Command	Description
mycel	Boot the server and open the web dashboard
mycel up	Start server (foreground); bootstraps ~/.mycel
mycel up -d	Start server (daemon)
mycel down	Stop server
mycel status	Show agent status
mycel doctor	Diagnose setup issues

Agents

Command	Description
mycel agent create [name]	Create agent with role and tool
mycel agent list	List all agents
mycel agent attach <agent>	Attach to agent's tmux session
mycel agent peek <agent>	Watch agent output
mycel agent send <agent> <msg>	Send message to agent
mycel agent stop <agent>	Stop agent
mycel agent delete <agent>	Delete agent

Channels

Command	Description
mycel channel list	List channels
mycel channel create <name>	Create channel
mycel channel send <ch> <msg>	Send message
mycel channel history <ch>	View history

Costs

Command	Description
mycel cost show	Cost summary (computed from provider session logs)
mycel cost budget show	Budget status
mycel cost daily	Daily spend series

Configuration

Command	Description
mycel config show	Show configuration
mycel config set <key> <val>	Set config value
mycel secret set <name>	Store encrypted secret
mycel tool list	List available tools
mycel mcp list	List MCP servers
mycel template list	List agent templates

Full reference: mycel --help or mycel <command> --help

Architecture

┌─────────────────────────────────────────────────┐
│                  mycel binary                    │
│                                                  │
│  ┌──────────┐  ┌───────────────────────────┐    │
│  │   CLI    │  │  HTTP Server              │    │
│  │ (Cobra)  │  │  + embedded Web UI        │    │
│  │          │  │  + MCP + SSE + Apps       │    │
│  └────┬─────┘  └────────────┬──────────────┘    │
│       │                     │                    │
│  ┌────┴─────────────────────┴─────────────────┐  │
│  │              pkg/ (Go packages)            │  │
│  │  agent · app · cost · gateway · mcp        │  │
│  │  provider · runtime · secret · home        │  │
│  └────────────────────┬───────────────────────┘  │
│                       │                          │
│  ┌────────────────────┴───────────────────────┐  │
│  │           Agent Runtimes                   │  │
│  │     tmux sessions │ Docker containers      │  │
│  │     git worktrees │ ~/.mycel entity state  │  │
│  └────────────────────────────────────────────┘  │
└─────────────────────────────────────────────────┘
                        │
            ┌───────────┼───────────┐
            │           │           │
         Claude      Gemini     Cursor ...

• Single binary — CLI, server, web UI, MCP all in one
• SQLite storage — zero external dependencies
• Agent isolation — each agent gets its own git worktree and tmux/Docker session
• Apps — 28 plugin integrations (Slack, Telegram, Discord, GitHub, …) with credentials in an encrypted vault

Configuration

Everything lives under ~/.mycel/ — your repos stay pristine. Config is one file, ~/.mycel/prefs.json:

{
  "version": 3,
  "server": {
    "host": "127.0.0.1",
    "port": 9374
  },
  "providers": {
    "default": "claude"
  },
  "runtime": {
    "default": "docker"
  },
  "storage": {
    "default": "sqlite"
  }
}

Vision

mycel treats agents as team members, not scripts:

• Dedicated worktree — an isolated git workspace per agent, so parallel work never conflicts
• Defined role — engineer, manager, reviewer, etc., with capabilities scoped to the role
• Structured collaboration — agents communicate through channels, not ad hoc chat
• Human oversight — you set budgets, define roles, review and merge PRs, and monitor activity from the CLI, web dashboard, or desktop app

Feature	mycel	Single-agent tools
Multiple parallel agents	Yes	No
Role-based hierarchy	Yes	No
Inter-agent communication	Yes	No
Git worktree isolation	Yes	No
Cost tracking per agent	Yes	Limited
Persistent agent memory	Yes	Session-only

Design principles: isolation by default, explicit (logged) inter-agent communication, cost awareness on every action, human-in-the-loop for critical decisions, and a CLI-first architecture with no infrastructure to stand up.

Roadmap: multi-repository orchestration, deeper CI/CD pipeline integrations, cost optimization recommendations, and (longer-term) distributed agent execution, cross-team agent collaboration, and enterprise features (SSO, audit logs).

Development

make build            # Build everything
make test             # Run all tests
make lint             # Run linters
make check            # Full quality gate
make run-mycel        # Run from source
make run-web          # Web UI dev server (hot reload)

See make help for all targets.

Contributing

Contributions welcome. Please run make check before submitting PRs.

License

MIT