Agent command center

Every AI agent on your machine, one coordination layer. Dispatch work from the web to whichever harness fits — Claude Code, Codex, Gemini, Cursor, OpenCode — and get a receipt for everything they do.

PHEWSH isn't another agent. It's the layer your agents plug into.

Your machine, right now

Checking for your local bridge…

Recent receipts

Receipts appear here when the bridge is running — the proof trail of what your agents actually did, read live from ~/.phewsh/ on your machine. Nothing leaves your machine; this page just renders it.

Connect in three steps

Install the CLI. curl -fsSL phewsh.com/install.sh | sh — no account needed.
Start the bridge. phewsh serve executes tasks instantly through any agent CLI you have installed. phewsh mcp setup goes deeper — it gives Claude Code, Cursor, and any MCP client a live session with your project's intent, constraints, and task queue.
Dispatch from anywhere. Open phewsh.com/intent → Work. The Live pill turns on, your installed agents appear, and every run lands in your receipts.

How the planes fit together

/cli — command
Author intent, sync it everywhere, run the bridges. The terminal surface.
/mcp — coordination
This page. Which agents are connected, what they're doing, with receipts.
/api — inference
One OpenAI-compatible key for any model, bounded by your project's budget.

Everything on this page is local-first: the bridge binds to localhost only, and receipts are plain JSON files you can open yourself. Cross-machine sync is on the roadmap. See /connect for the full integration picture.