/

AI provider settings

AI provider settings

The AI provider settings page picks which AI powers Draft with AI on the Skills page and Generate from document. Two options are supported: the local Claude Agent SDK, or any OpenAI-compatible endpoint.

What you'll do

  • Choose Claude Agent SDK or an OpenAI-compatible provider.
  • Configure the base URL, model, and API key.
  • Test the connection.

The two providers

Claude Agent SDK

  • Uses the local claude binary already installed and signed in on the server.
  • No API key is required.
  • Best for teams who already run Claude locally and want zero extra configuration.

OpenAI-compatible

  • Works with any endpoint that speaks the OpenAI Chat Completions API. That includes OpenAI itself, Ollama, LM Studio, Together, and Anthropic-via-proxy setups.
  • Configure the base URL, the model name, and an API key if the endpoint requires one.

Step-by-step

  1. Open AI provider settings from the top-level Settings section.
  2. Pick a provider using the radio group at the top.
  3. Fill in the fields for the chosen provider:
    • Base URL (OpenAI-compatible only).
    • Model.
    • API key (optional; leave blank for endpoints that do not require it).
  4. Click Test connection. A success message means the app can reach the endpoint and get a valid response.
  5. (OpenAI-compatible only) Click Load models to fetch the model list from your endpoint. Pick one and it fills the Model field.
  6. Click Save.

Good to know

  • Your API key stays on the server. The browser only sees whether a key is configured, not the value.
  • Two features depend on this configuration: Draft with AI on the New skill page, and Generate from document. Both will refuse to run with a "not configured" message if you have not saved a provider yet.

Troubleshooting

  • Claude is unavailable, "not installed", the claude binary is not present on the server. Install it or switch to an OpenAI-compatible provider.
  • Claude is unavailable, "authentication needed", sign in with claude on the server (usually by running it interactively once).
  • OpenAI-compatible test fails with a network error, the base URL cannot be reached. Verify the URL and any firewalls between the app and the endpoint.
  • "Invalid AI endpoint, cannot reach private, loopback, or link-local hosts", the app blocks AI base URLs that resolve to loopback, private, link-local, or cloud metadata addresses to prevent SSRF. In development, plain http:// is permitted so Ollama and LM Studio still work; production requires https:// and a publicly resolvable host.
  • The Save button stays greyed out even though I picked a provider, edit any field or re-select the radio to mark the form as dirty, then Save.
  • "AI not configured" banner won't clear after saving, refresh the page.
PreviousDev drawer and Copy as cURLNext API reference viewer