Coding agents got good. Managing them didn't. I open-sourced the control room I built for it (MIT)

Run more than one coding agent and your day becomes five terminals, no shared memory, and git log as your only audit trail. Every vendor's answer is "use only ours." o8 is the layer above all of them. One orchestrator scopes the work and dispatches it, every worker runs in its own git worktree, and nothing merges until you approve it. Claude Code, Codex, Gemini, Aider, Goose and eight more run behind a single adapter contract, so you're not picking a lab and living inside their box. What's in it: Merge gate — rebase, typecheck, head-SHA lock, ordered merge. Workers can't merge their own work; execution and approval are separate authorities by construction. Engineering Brain — ask about your repo or your fleet's history, get answers with citations. "What did the agents ship yesterday" is a query, not archaeology. Organizational memory — repeated fixes come back as proposed rules you accept or reject. Replayable audit trail — every packet, merge and decision is a row in SQLite on your own disk. No telemetry; it's off by default and opt-in. Mobile — approve or reject a merge from your phone while the fleet keeps working. Canvas — the IDE as a work surface, with agents, diffs and a live preview side by side. Free, MIT, and it runs on the subscriptions you already pay for. macOS today; Windows and Linux are mapped and in progress. ~88MB, signed build, auto-updates. I built it, and I used it to build itself — 5,500+ commits, most of them dispatched through it. Issues are open, and adding a runtime is a documented patch — the adapter contract is one file. If there's a CLI you want in there, that's the shortest path to a merged PR. What are you using right now to stop two agents from stepping on each other?