A simple, fully browser-based LLM completion client featuring:

• integration to multiple providers and models
• chat history via InstantDB

Try it here: https://betafunk.dk/tjat

Q: Does this app steal my API keys?

A: No, keys are stored locally, and only submitted to model providers

Q: How do I save my chat history?

A: Go to InstantDB, create a new app, and copy the app-id into Tjat's settings.

Q: Can people read my chat history?

A: Only if they can guess your InstantDB app-id!

Q: Inserting my AWS / Cloudflare keys into the browser, are you nuts?

A: Still only stored locally in the browser.

To make Ollama accept requests from the browser, you need to allow additional origins when starting Ollama, eg:

$ OLLAMA_ORIGINS=https://betafunk.dk ollama serve

and then start your model, eg:

$ ollama run qwen3:8b

To be able to reproduce a request with attachments, you can add credentials for R2 or S3 in the settings.

Once enabled, attachments will be uploaded to the configured bucket.

You'll need to add CORS rules to your bucket to allow access from wherever you've deployed Tjat. The necessary rules look like this for Cloudflare R2:

[
  {
    "AllowedOrigins": [
      "https://mytjattywebsite.ay"
    ],
    "AllowedMethods": [
      "GET",
      "PUT",
      "HEAD"
    ],
    "AllowedHeaders": [
      "Content-Type"
    ]
  }
]

The randomllm service provides a mock LLM API for testing purposes.

cd randomllm
source venv/bin/activate
pip install -r requirements.txt
python randomllm.py

The service will run on port 8042 by default, or you can specify a port:

python randomllm.py 8080

cd app
npm i

$ npx shadow-cljs server

In a repl:

(shadow/watch :app)

Visit http://localhost:8000

Check out the deploy.sh script.