CO_DE Code in your w_Ai Get CO_DE today
macOS · Apple Silicon · free

Your AI tools. One native workspace. Integrated coder and orchestrator.

Claude Code, Codex, Cursor Agent, Kimi Code — side by side in one structured workspace. The coder and the orchestrator come with the app, built on Pi. Over MCP the same seat is open to anyone else: any CLI, inside the app or outside it, and any local model in any harness. The brain behind it is yours — your own GPU, or your Claude or Codex subscription through that vendor's own CLI.

Your providers. Your subscriptions. Your orchestrator. No new subscription required.

CO_DE workspace with multiple AI agents, terminals, and a browser open side-by-side

02   Workflow

Browser. Terminal. Tasks.
All in one desk.

Work side-by-side — route prompts to any agent, keep your browser, terminal, and task context together in one structured workspace. Nothing closes when you switch focus.

CO_DE workspace showing a browser pane, terminal, and task/context panel running side-by-side

03   The orchestrator seat

The seat is yours.
And it runs off the meter.

Pick the brain that runs the meeting: a local model on your own GPU, your Claude or ChatGPT subscription through its own official CLI, or any OpenAI-compatible / LiteLLM endpoint — your login, your terminal. It dispatches to the coding agents you already pay for, each on its own worktree, and answers them when they get stuck. The seat is not locked to the app's own chat either: over MCP any CLI — internal or external — and any local model in any harness can sit in it. CO_DE detects the CLIs and local servers already on your machine; Settings → Local Tools shows what is ready.

Local model Your own CLI login OpenAI-compatible
CO_DE Settings → Local Tools panel showing detected local CLIs and models

04   Features

Every tab,
and what it does.

What is in the app today, tab by tab. The list is kept in the repo next to the code and re-read at every release, so it describes the build you download — not a plan.

CO_DE with sixteen agent panes working side by side — Claude Code, Codex, GPT-5.6, Opus and local models — with the agent sidebar on the left and the steer and report rail on the right
Sixteen agents on one wall: each pane a real CLI with its own login, its own worktree and its own face — the sidebar on the left, what each one asked for on the right.
Two agent panes in CO_DE: Kimi Code running on a local GGUF model with the orchestrator MCP connected, and Claude Code on Opus 5
The real CLIs, in panes the app ownsKimi Code on a local model, Claude Code on your subscription — each with the status its own session reports.
A CO_DE browser pane with the AI Full pill in the address bar and Design mode armed
A browser with a leashThe pill says what a model may do in this pane — Off, Read only, Full access — and Design mode turns a click into a note.
The Scenarios canvas in CO_DE: an Orchestrator node wired to coordinator nodes, each with its own model
Scenarios, on a canvasAn Orchestrator at the root, a coordinator per node with its own model, and a pass that walks the wires on its own.
CO_DE sidebar showing desks, branches and agents, each agent with its own face
Every agent has a faceColour says what it is doing, each session has its own temperament, and one switch puts the provider logos back.

05   Changes

What shipped,
release by release.

Straight from the app's own history, newest first — rewritten with every build that goes out.

Get CO_DE today

macOS available now. Windows and Linux later. Use your own CLI subscriptions — no central sync required.

Windows
Coming soon
Planned via GitHub Releases
Linux
Coming soon · AppImage
Planned via GitHub Releases
Quick setup

Install the local provider CLIs you want to use:

npm install -g \ @anthropic-ai/claude-code \ @openai/codex \ @google/gemini-cli \ @cursor/sdk

Already have them installed? CO_DE will detect them automatically.

After installing new CLIs: Settings → Refresh Local Tools.

Using Ollama? Install the desktop app — CO_DE detects it via localhost.