$ spawn --team research-squad

The AI-native
agent orchestration
IDE.

Spawn teams of AI coding agents across multiple models and providers. Watch them work in floating terminal windows. Orchestrate from above.

Free · Open source · Works with Claude, Codex, Kimi, Gemini, Grok, Copilot, Pi · 200+ models via OpenClaude
Spawning a full team of agents in The Cog
Spawning an orchestrator + 6 workers + 2 reviewers + a researcher in seconds
The gist

The Cog is an Electron IDE built from the ground up for AI agents as primary workers. You open a project, spawn a team of agents across different models and providers, and they coordinate through 31 MCP tools — messaging, task boards, shared knowledge, scheduling, a sketchpad they can see, file operations — all in one workspace. You steer from above.

What's inside

Build a crew, ship faster.

Per-agent color themes

Per-agent color themes

Right-click any terminal to customize chrome, border, background, and text colors. 8 presets (Sunshine, Ocean, Crimson…) or roll your own. "Apply theme by role" colors your whole team in one click.

Community Teams browser

Community Teams

Share your best team setups with the world. Browse, star, and import configurations from other users. Categories for research, coding, review, full-stack, decomp. No account needed.

Remote View on phone

Remote View & Workshop

Check on your workshop from your phone via a Cloudflare tunnel. Unlock Workshop mode with a passcode and get a spatial canvas mirroring your desktop — pinch, zoom, pan, kill agents remotely.

┌──────────────────────────────┐
│ [orchestrator] Opus          │
│  ↓ send_message              │
│  ↓ post_task(target_role)    │
├──────────────────────────────┤
│ [worker-1]  Sonnet           │
│ [worker-2]  Sonnet           │
│ [reviewer]  GPT-5.5          │
│ [researcher] Kimi K2.6       │
└──────────────────────────────┘

Multi-model teams

Mix and match. Claude orchestrates while Kimi researches and Codex reviews. 31 MCP tools let agents message, share, assign tasks, schedule themselves, and read each other's terminals. Agents never poll — auto-nudge on events, zero wasted tokens.

┌─ Sketchpad ───── page 1/3 ─┐
│   ┌──────┐   "fix the      │
│   │ map  │    spawn flow"  │
│   └──────┘       ╲         │
│      ✎ ─────────▶ ◯ note   │
│                            │
│  agent → view_sketchpad    │
│         _page(1)  👁         │
└────────────────────────────┘

A Sketchpad your agents can see

A per-project canvas — sticky notes, text, photos, and freehand drawing across pages. The twist: agents read your pages with vision via view_sketchpad_page. Sketch a map, a layout, or a plan and your crew works from what you drew.

┌─ Schedules ────────────────┐
│ every 45m → "keep going"   │
│ every 2h  → "commit work"  │
│ ▶ running    ⏸ 7h 12m left │
│ ───────────────────────────│
│ agent → schedule_prompt()  │
│   → 📥 inbox: approve?      │
└────────────────────────────┘

Keep going while you're away

Fire recurring prompts at any agent on a schedule — leave a "keep going" nudge and come home to a team that never went idle. Agents can even schedule themselves: gated proposals land in your inbox, or flip a project to fully autonomous.

Supported CLIs
Claude Code
Codex CLI
Kimi CLI
Gemini CLI
OpenClaude
Copilot CLI
Grok CLI
Pi (pi.dev)
Plain Terminal

200+ models via OpenClaude (OpenAI-compatible providers: GPT, DeepSeek, Ollama, Qwen, Mistral, and more).

Get started

Clone and run.

terminal
$ git clone https://github.com/the-cog-dev/cog.git
$ cd cog
$ npm install
$ npm run dev
Requires: Node.js 20+, at least one AI CLI installed (Claude Code, Codex, Kimi, Gemini, etc.)
Coming soon One-click installer (.exe / .dmg / .AppImage) — skip the clone.