kunji v0.1.2

What's new

• Menu: run kunji and use the arrow keys. It stays open until you exit, so you don't need to type commands.
• Saved prompts: kunji @commit writes commit messages from your staged changes. @explain, @review, @fix, @summarize and @translate are built in, and you can add your own.
• Chat history: every chat is saved on your computer. Continue the last one with kunji chat -c, or find old ones with kunji history search.
• Attach files: /file app.py in a chat, or kunji ask -f app.py "what does this do?".
• Easy upgrades: kunji tells you when a new version is out, and kunji upgrade installs it.
• Website: https://testwithsuraj.github.io/kunji/

Install or upgrade

pip install --upgrade kunji

If pip isn't found, use python -m pip install --upgrade kunji.
No Python on Windows? Download kunji-windows-x64.exe below.

Then run kunji to get started. Full command guide: docs/COMMANDS.md

⭐ Star kunji · 🐛 Report a problem

kunji v0.1.0

kunji 0.1.0 — first release 🔑

Your keys. Every AI model. One terminal.

kunji is a free, open-source chat app for your terminal. Bring your own API key from any
provider, switch models in the middle of a conversation, and see what every reply costs.
Your keys stay in your computer's keychain, and your prompts go straight to the provider.

Install

System	How
Windows	pipx install kunji, or download kunji-windows-x64.exe below (no Python needed)
macOS	pipx install kunji
Linux	pipx install kunji, or download kunji-linux-x64 below

Then add a key and start chatting:

kunji keys add
kunji

What you can do

• Use 10 providers from one place: OpenAI, Anthropic, Google Gemini, Groq, OpenRouter,
  DeepSeek, Mistral, xAI, Together AI and Ollama (free, local), plus any
  OpenAI-compatible server such as LM Studio.
• Switch models without losing the conversation: /model groq:llama-3.3-70b-versatile
• See what every reply costs, with /cost for the session and kunji usage for the week,
  in your own currency (₹ supported).
• Set daily budgets. kunji warns at 80% and stops at 100%, so there are no surprise bills.
• Compare models side by side: kunji compare "your prompt" shows which was fastest
  and cheapest.
• Keep working when a provider is down. kunji falls back to your next model.
• Pipe anything in: npm test 2>&1 | kunji ask "why does this fail?"
• Keep your keys safe in Windows Credential Manager, macOS Keychain or Linux Secret
  Service. Never in a text file, and kunji checks each key before saving it.
• Looks good in any window size: a clean, minimal design that adapts as you resize.

Privacy

kunji has no account, no analytics and no server in the middle. The only automatic network
calls are a daily version check and a public price list, and you can turn both off.
Details: Privacy policy

Feedback

This is the first release, so your feedback shapes what comes next.
⭐ Star the repo if it's useful · 🐛 Report a problem · 💡 Ideas welcome

Coming next: saved prompts (kunji @commit), chat history, attaching files, and
winget install for Windows.

Downloads are listed below. Check them against SHA256SUMS.txt.
Full changelog: CHANGELOG.md