MadAppGang/claudish

🔮 Claudish – A universal proxy for Claude Code

What it is

• A command‑line interface (CLI) that lets you run Claude Code (Anthropic’s coding assistant) with any LLM you have access to. It works by starting a local server that mimics the Anthropic API, then forwards the request to the provider you specify.

Why it matters

• You can keep a single workflow (Claude Code’s powerful prompt language, file‑based instructions, sub‑agent delegation) while swapping between models from Anthropic, OpenAI, Google Gemini, Ollama, OpenRouter, and dozens of other vendors. This saves you from maintaining multiple CLI tools or paying for several subscriptions.

Key capabilities

• Multi‑provider routing – Use a provider@model[:concurrency] syntax (e.g. g@gemini-3-pro-preview, ollama@qwen3-coder:latest). Supports cloud APIs, local servers, and enterprise services like Vertex AI.
• BYOK (Bring‑Your‑Own‑Key) – No extra fees; just set the appropriate environment variable or let the interactive prompt collect your API key.
• Full Claude Code compatibility – All Claude Code flags (--agent, --effort, --permission-mode, …) are passed through unchanged, so existing Claude Code scripts keep working.
• Offline mode – Run entirely on a local model (Ollama, LM Studio, vLLM, MLX) so your code never leaves the machine.
• Vision proxy – Non‑vision models automatically get image descriptions from Claude, giving every model “sight”.
• Developer ergonomics – Interactive setup, auto‑detection of models, JSON output for tool integration, cost‑tracking, and a monitor mode that logs traffic for debugging.
• Cross‑platform – Works with Node.js or Bun, and can be run via npx/bunx without a global install.

Installation

# Quick one‑liner (Linux/macOS)
curl -fsSL https://raw.githubusercontent.com/MadAppGang/claudish/main/install.sh | bash

# Homebrew (macOS)
brew tap MadAppGang/tap && brew install claudish

# npm / Bun
npm install -g claudish   # or  bun install -g claudish

You can also run it directly without installing:

npx claudish@latest --model openrouter@anthropic/claude-3.5-sonnet "refactor this function"

Typical workflow

1. Configure your API keys – set OPENROUTER_API_KEY, GEMINI_API_KEY, OPENAI_API_KEY, etc., or let claudish login handle OAuth for supported providers.
2. Pick a model – either let the interactive selector choose, or specify explicitly:

   claudish --model g@gemini-3-pro-preview "write a unit test"
   claudish --model ollama@qwen3-coder:latest "implement login flow"
   

3. Run Claude Code – any flags you would normally give to claude are accepted after claudish.
4. (Optional) Use in agents – the tool provides an AI‑agent guide (claudish --help-ai) and recommends a file‑based pattern to avoid polluting the LLM’s context window.

Configuration

• Environment variables are auto‑loaded from a .env file in the current directory.
• Settings such as default model, proxy port, concurrency limits, and telemetry can be overridden with CLAUDISH_* variables or a custom config file (--config <file>).

License

• MIT (see LICENSE).

Where to learn more

• Official website: https://claudish.com
• Full docs: https://github.com/MadAppGang/claudish/blob/main/docs/index.md
• Issue tracker for bugs & feature requests: https://github.com/MadAppGang/claudish/issues