┌─  kunji chat  ─────────────────────────────────────────────────┐
│  Provider     Model            Key         Budget        State │
│────────────────────────────────────────────────────────────────│
│  Anthropic    claude-sonnet    keychain    $0.38 / $2    READY │
└────────────────────────────────────────────────────────────────┘
enter send  ·  alt+enter newline  ·  / commands  ·  ctrl+d quit

› explain recursion in one line
  A function that solves a problem by calling itself on a smaller
  piece of it, until it reaches a case simple enough to answer.
  claude-sonnet · 18 in · 31 out · $0.0005 · 1.2s

› /model groq:llama-3.3-70b-versatile
✓ Switched to groq:llama-3.3-70b-versatile  conversation kept

Kunji is a terminal app for chatting with any AI model using your own API keys. You pay each provider directly. Kunji never sees your bill, your keys or your prompts.

Why kunji

• One tool, every provider. OpenAI, Anthropic, Google Gemini, Groq, OpenRouter, DeepSeek, Mistral, xAI, Together, Ollama (local), plus any OpenAI-compatible server.
• Switch models mid-conversation with /model — the conversation carries over.
• Live cost meter. Every reply shows tokens and cost. /cost and kunji usage show totals, in your currency.
• Daily budgets. Warns at 80%, blocks at 100%, per provider or overall.
• Compare mode. One prompt, several models, side by side — fastest and cheapest highlighted.
• Automatic fallback. Rate-limited or down? Kunji retries on your next model.
• Keys in your OS keychain (Windows Credential Manager, macOS Keychain, Secret Service), never in a plain-text file.
• Pipes. npm test 2>&1 | kunji ask "why is this failing?"
• Looks professional. Minimal monochrome design, responsive to any terminal size.

Install

Requires Python 3.11+.

pipx install kunji        # recommended
# or
uv tool install kunji
# or
python -m pip install --user kunji

winget, Scoop and Homebrew packages are on the roadmap.

Quick start

kunji keys add            # pick a provider, paste your key, choose a default model
kunji                     # start chatting

Free and local? Install Ollama, then kunji keys add ollama.

Commands

In chat

Providers

Any OpenAI-compatible server (LM Studio, vLLM, LiteLLM proxy…):

kunji providers add lmstudio --base-url http://localhost:1234/v1 --no-key

Security and privacy

• API keys are stored only in your operating system's credential store, or read from environment variables. They are sent only to the provider they belong to.
• Kunji's config file never contains keys. Errors are scrubbed so keys never appear on screen.
• Usage and cost history stay on your machine (kunji config path).
• Network calls kunji makes on its own: the public model price list (OpenRouter, plus this repo's catalog/pricing.json) and the PyPI version check. Neither sends anything about you. Turn them off with kunji config set network.catalog false and kunji config set updates.check false.
• Found a vulnerability? Please follow SECURITY.md.

Details: docs/PRIVACY.md.

Configuration

kunji config set display.currency INR
kunji config set display.usd_rate 84          # your conversion rate
kunji budget set all 200                      # ₹200/day across every provider
kunji config set fallback '["groq:llama-3.3-70b-versatile", "ollama:qwen2.5"]'

All options: docs/CONFIG.md.

Free and Pro

Everything that runs on your machine is free and open source, forever — chat, switching, keys, cost meter, budgets, compare, fallback, pipes. A future optional Pro plan will add cloud features that need a server (sync across devices, shared team keys with per-person budgets, a smart router). See ROADMAP.md.

Contributing

Bug reports, ideas and pull requests are welcome — start with CONTRIBUTING.md. Architecture overview: docs/ARCHITECTURE.md.

License

Apache-2.0. "kunji" and the kunji logo are not covered by the license; please use a different name for forks.