Deep Claude

deep-claude 🐋

  Drive Claude Code with any model.
  OpenRouter by default — Gemini, GPT, DeepSeek, Grok, Qwen, Claude, and more — or any
  Anthropic-compatible endpoint you point it at. Same harness, a different mind, cleanly isolated.

  Website ·
  Install ·
  Usage ·
  Endpoints ·
  How it works ·
  Troubleshooting

  deep-claude points Claude Code at
  OpenRouter through a tiny local proxy and redirects its state
  into a private home, so your real Anthropic login is never touched. Pick the models you want, then
  type deep-claude followed by any normal claude arguments — you get the full
  agentic harness (tools, MCP, slash commands, sub-agents) running on whichever model you chose.

Why?

Claude Code is one of the best agentic coding harnesses there is: the planning loop, tool use, MCP
servers, sub-agents, permissions, and slash commands are all _harness_, not _model_. Lots of models
speak (or can be made to speak) the Anthropic API — so you can keep every bit of that machinery and
swap the brain.

- One key, any model. An OpenRouter key gets you Gemini, GPT, DeepSeek, Grok, Qwen, Claude, and
  hundreds more behind a single endpoint.
- A curated, pretty picker. deep-claude pick is a searchable, multi-select TUI over the whole
  catalog (with context windows and pricing), and assigns your picks to Claude Code's switch slots so
  you can flip between them in-session. It's a two-level provider picker: the first screen lists
  every provider (OpenRouter plus each personal endpoint) — set a provider's API key with e, add a
  new one with ＋ Add provider, then press ↵ to drill into that provider's models. The OpenRouter
  key itself lives here too, so one screen manages every key. ⌃S saves everything.
- Real multi-model workflows. A sub-agent's model: accepts a full…