Everything kunji does

Every model, one place

Ten providers are built in, and any OpenAI-compatible server works too. Add a key once and kunji checks it before saving.

• OpenAI, Anthropic, Google Gemini, Groq, OpenRouter, DeepSeek, Mistral, xAI, Together AI, Ollama
• LM Studio, vLLM, LiteLLM or your company proxy with kunji providers add
• Live model lists, straight from each provider

Provider setup guides

$ kunji keys add groq
✓ Verifying with Groq  0.4s  21 models available
✓ Saving key to Windows Credential Manager  0.1s
◆ Default model (type to search) llama-3.3-70b-versatile

Switch models without starting over

Ask one model, then ask another to check or improve the answer. The whole conversation carries over.

• /model with Tab completion for model names
• A default model per provider, and one overall

Switch and pick models

› /model anthropic:claude-sonnet-4-5
✓ Switched to anthropic:claude-sonnet-4-5  conversation kept
› review the function you wrote above

Know what it costs, and set a limit

Every answer shows its tokens and price. Daily budgets warn you at 80% and stop at 100%, per provider or across all of them, in your own currency.

• /cost in a chat, kunji usage for the week or month
• kunji budget set all 500 with costs in ₹, $, € or any currency

Costs and budgets

  gpt-4o · 1,204 in · 380 out · ₹0.79 · 3.4s

! 80% of today's budget used for OpenAI: ₹161.40 / ₹200.00

Compare models side by side

Send one prompt to several models at the same time, then see each answer and which was fastest and cheapest.

• Find the cheapest model that's good enough for a task
• Double-check important answers

Compare models

$ kunji compare "explain recursion" -m openai:gpt-4o-mini -m groq:llama-3.3-70b-versatile
✓ openai:gpt-4o-mini  1.9s · 31 tok
✓ groq:llama-3.3-70b-versatile  0.6s · 27 tok
  groq:llama-3.3-70b   0.6s   fastest cheapest

Saved prompts for daily work

Instructions you use often, run in one word. Six are built in, and you can write your own as simple Markdown files.

• @commit writes a commit message from your staged changes
• @explain, @review, @fix, @summarize, @translate
• Use them from the command line or inside a chat

Saved prompts

$ git add . && kunji @commit
  Add password reset by email

  Users can now request a reset link from the
  login page. Links expire after 30 minutes.

Chats you can come back to

Every conversation is saved on your computer. Pick up tomorrow exactly where you left off, or search old answers.

• kunji chat -c continues the last chat
• kunji history search docker finds old ones
• Export any chat as Markdown

Chat history

$ kunji history
  ID       Title                         Updated
  3fa2c1   fix the login redirect bug    03 Nov 22:10
  9b07de   compare postgres vs sqlite    02 Nov 18:44

Files and pipes

Attach code and logs, or pipe any command's output straight into a model. Plain output when you need it for scripts.

• /file app.py in a chat, with path completion
• kunji ask -f app.py "what does this do?"
• --raw output for scripts and CI

Pipes and scripts

$ npm test 2>&1 | kunji ask "why does this fail?"
  The test expects a 200 but the route returns 302
  because the session cookie isn't set in the test…

A menu when you don't want to type commands

Run kunji on its own for an arrow-key menu that stays open until you exit. Every action is also a command for scripts.

The menu

◆ What would you like to do?
 ❯ Start a new chat          anthropic:claude-sonnet
   Continue last chat        fix the login redirect…
   Run a saved prompt        @explain, @commit, …
   Compare models
   Usage and budgets

Free local models

With Ollama, models run on your own computer: no key, no bill, and it works offline. Mix local and cloud models in the same chat.

Free local models with Ollama

$ ollama pull llama3.2
$ kunji keys add ollama
$ kunji chat -m ollama:llama3.2

Private by design

Keys live in your OS keychain and go only to their own provider. Prompts go straight to the provider you choose. There's no kunji server, no account and no analytics.

• Keys always shown masked, and removed from error messages
• Chats, prompts and usage stay on your computer
• Open source under Apache-2.0, so anyone can check

Privacy and security

$ kunji keys list
  Provider    Key                 Status
  anthropic   sk-ant••••••••3f9a  KEY SET
  groq        gsk_ab••••••••91c2  KEY SET