Co-Vibe — BLUF#

Bottom line: Co-Vibe is the coordination layer for teams running multiple AI coding agents like Claude Code, Codex, and Cursor, plus Paxel profiling. It makes parallel agent work visible, blocks real duplicate work before it becomes a commit, captures verified (deterministic, not self-reported) agent metrics, and shows them in developer/team profiles — coordination, not surveillance.

What it does (one line each)#

Sees the agents. Claude Code via native hooks + local transcript usage-counter parsing; Codex/Cursor via MCP sessions + counter-only telemetry; every signal carries provenance.
Stops duplicate work. A task can't start over a genuine overlap without a human reason (100% recall on real duplicates, >90% precision); the agent gets the warning over MCP, not just the UI.
Tracks exact changes (file paths, diff hashes, branch, commit SHA) without storing code.
Profiles builders with the official Paxel tool, run locally from the developer's own machine.
Self-hosts. The trust boundary — local companion, hooks, redaction — is open-source; the server is multi-tenant ready.

90-second path#

npm install && npm run setup && npm run dev → open the printed UI URL.
Settings → Agent connection → copy the companion install + covibe-local setup, then approve the browser device prompt.
Start Claude Code and Codex on the same repo.
Watch active sessions, files changed, and an overlap warning appear; open Performance for verified usage metrics.
Run Paxel from your profile, then covibe-local paxel on your machine (Paxel is strictly user-triggered), for a builder report.

Where to go next#

Install/run/auth: README.md
The eight MCP tools + hard rules: docs/engineering/mcp-contract.md
Why duplicate-work detection is trustworthy: docs/engineering/overlap-precision.md
How agent metrics stay honest: docs/engineering/verified-agent-telemetry.md
Running the official profiler locally: docs/engineering/paxel-integration.md
Machine-readable summary for AI search: /llms.txt