JSLab feature notes: one docs page, explained. Today: AI providers and keys.
JSLab's Ask AI panel answers questions about the code in your tab and returns changes as a diff. It doesn't come with a model of its own: it uses your own key, or Ollama locally. This note is about that setup.
One provider at a time
Everything lives in Settings → AI. Provider starts at None, which turns AI Chat off. The other choices:
| Provider | Needs an API key | Standard endpoint |
|---|---|---|
| Ollama (Local) | No | http://localhost:11434 |
| OpenAI | Yes | https://api.openai.com/v1 |
| Anthropic | Yes | https://api.anthropic.com |
| Google Gemini | Yes | https://generativelanguage.googleapis.com/v1beta |
Ollama is a separate, free program that runs models on your own Mac. Install and start it yourself, choose Ollama (Local), and you're done. No key.
Here's AI Chat in a real JSLab tab with Ollama as the provider: the provider shows as a chip at the top of the panel, and the question goes to the model on your own Mac.
Picking a model
Each provider has its own Model field. Leave it blank for the default that ships with this version of JSLab, or type the name you want. The field also lists the models the provider reports; for Ollama, that's the models you've installed. Refresh fetches the list again.
There's a Base URL per provider too. Blank means the standard endpoint above; set it to point at, say, an Ollama server on another machine.
Where the key lives
The API Keys section of Settings → AI has a field for OpenAI, Anthropic and Google Gemini. Paste a key, click Save, and the field shows A key is stored. Remove deletes it.
The details that matter:
- Keys are stored in your macOS Keychain, never in
settings.json. - JSLab sends a key only with requests to that provider's Base URL, and only in a request header, never in a URL.
- Requests go straight from your Mac to the provider, with your own key.
- Without a stored key, a request stops before it's sent and AI Chat reports a credentials error.
What gets sent
Each message carries JSLab's instructions, the current tab's code (up to 100 KB), the conversation so far, and your message. With Send Recent Output on (the default), the last run's output goes too, up to 20 KB. Turn it off to send only code and conversation. Nothing else from your Mac is sent, and with Ollama at its standard local address, none of it leaves your Mac.
JSLab launches Nov 15. The full page: https://js-lab.app/docs/ai-providers
Top comments (0)