Separate identities for Claude, Codex & Antigravity

Run several agents on one machine and they blur into one anonymous stream. Paigy names them: each agent gets its own identity, its own voice on calls, and its own leash. Running agents in parallel multiplies the needs-input moments — this is how one person answers all of them.

How it works

paigy-harness setup walks the machine, finds the agents you already have, and gives each one a dedicated identity slot — mcp-agent for Claude Code, codex for Codex, antigravity for Antigravity. No shared tokens, no crossed wires; agents you don't have are skipped, never installed. Grok is detected by the same installer. Each Grok conversation is its own agent, from the session Grok already names, so two chats do not share one identity.

• Its own name and voice. Pick a voice per agent in the app — on a shared call you hear which agent is talking.
• Its own limits. Cap one agent at banners while another may ring — each identity has its own reach.
• Its own off switch. Disconnect one agent from the app at any time without touching the rest.

Set it up

Re-running is always safe — every step skips itself when already done, so it doubles as "add another agent".

Wiring an MCP config by hand instead? Set PAIGY_AGENT in each client's config so every tool keeps its own identity.

Works with: Claude Code, OpenAI Codex, Grok, Google Antigravity, and any MCP-compatible client — Cline, Continue, Zed, Goose, and local-model setups included.