AI Collab organizes your local AI coding tools into a team with memory, collaboration, and recoverability. A native macOS app and CLI managing persistent multi-AI Scenarios — isolated workspaces, typed agent-to-agent messaging, one-click resume. Open source under MIT.

Four agents across two Scenarios messaging each other through the Host, without disturbing each other
A team with memory.
Isolated workspaces, reliable messaging, recoverable state — the primitives multi-AI work actually needs.
Each Scenario is a long-lived work room that keeps its workspace, participants, and conversation state. Close it tonight, resume it tomorrow with one click.

Every Scenario gets its own multi-repo workspace. Run many experiments in parallel without directory conflicts or state pollution.

Mix Claude, Codex, and locally hosted models in one Scenario, each with its own role — implementer, reviewer, analyst.
Agents exchange typed messages — questions, review requests, pushback — through a single Host authority with delivery receipts.
A generic adapter protocol: declare your repositories and environment in config files, or let the app scan your directory and draft them for you.
The macOS app is the primary entry for daily work; the CLI drives automation and diagnostics against the same Host.
Open source under MIT
Native macOS app — no Electron
Your code never leaves your machine
Vendor-neutral: Claude, Codex, local models
Recoverable after restart or interruption
Battle-tested on our own research daily
AI Collab processes everything on your device. Zero data collection. No account required.