Changelog

Every release with its changes, newest first. Downloads are on the GitHub releases page.

To update kunji, run kunji upgrade (or pip install --upgrade kunji).

All notable changes are listed here. The format follows Keep a Changelog and the project uses semantic versioning.

[Unreleased]

Documentation site at https://testwithsuraj.github.io/kunji/docs/ with guides, provider setup pages, a full command reference generated from the code, troubleshooting and FAQ.
New website: interactive terminal demos (chat, switch model, compare, @commit, budget), a Features page, light/dark theme (shared with the docs), Hindi pages and a mobile menu.
Demo GIF and videos in the README, recorded from the real CLI (scripts/demo/).
How-to guides in the docs: Claude in the terminal, a free AI CLI for Windows, the Groq API from the command line, ChatGPT API vs Plus cost, and AI commit messages.

First release with these features (0.1.1 was built but never published).

A menu, so you don’t have to remember commands. Run kunji and pick what you want with the arrow keys: chat, saved chats, prompts, compare, models, providers, usage, budgets and settings. It stays open until you choose Exit, and leaving a chat brings you back to it. First time? It starts by helping you add a provider.
Saved prompts. Ready-made instructions you run in one word: kunji @commit writes a commit message from your staged git changes, and @explain, @review, @fix, @summarize and @translate are built in. Use them inside a chat too (@explain this error …), or create your own with kunji prompts add.
Chats you can come back to. Every chat is saved on your computer. kunji chat -c continues the last one, kunji history lists them all, and kunji history search finds old answers.
Attach files. /file app.py in a chat (Tab completes the path), or kunji ask -f app.py "what does this do?". Text files up to 200 KB each.
Upgrade notices and kunji upgrade. When a new version is out, kunji tells you at startup and upgrades itself with one command.
Website at https://testwithsuraj.github.io/kunji/ with install steps, privacy policy and terms.

Interactive chat with streaming markdown answers, slash commands and history.
Providers: OpenAI, Anthropic, Gemini, Groq, OpenRouter, DeepSeek, Mistral, xAI, Together, Ollama, and custom OpenAI-compatible servers.
/model switching mid-conversation with tab completion.
API keys in the OS keychain (or environment variables), verified before saving.
Live cost meter, kunji usage, and daily budgets (warn at 80%, block at 100%).
kunji compare — one prompt, several models, with fastest/cheapest summary.
Automatic fallback to the next model on rate limits or outages.
kunji ask with pipe support and --raw output for scripts.
kunji doctor, kunji models, kunji providers, kunji config.
Responsive, minimal theme that adapts to any terminal width.
Update notice via PyPI and signed (Ed25519) cloud check-in client (off until a server URL is configured).

Standalone downloads for Windows (kunji-windows-x64.exe) and Linux; macOS users install with pipx install kunji (a macOS download can be switched on later).
Releases run automatically when the version number changes, with notes taken from this changelog.

Compact one-row summary tables (Strip) for the chat header, /cost, keys add, usage totals and version — 5 lines instead of ~11, more room for the conversation.
kunji compare summary is now a table: time, tokens, cost and fastest/cheapest badges for every model.
README badges restyled for readability on light and dark GitHub themes.