AI Provider
Connect your workspace’s own AI key so the AI features run on it instead of the shared quota — and so the features that only run on your own key run at all.
By default, Meldoc uses a shared AI quota for Ask AI and writing features. A workspace admin can connect a provider on the AI Provider Settings page: one provider is active at a time, and the key is used for content generation across the workspace.
Search embeddings always run on Meldoc’s own key, whatever you configure; every other AI feature asks your provider once one is configured. The documentation review on pull requests goes the other way: its semantic pass runs only with your own key, and without one it reports dead references and says in the check summary that it could not judge meaning. See Documentation Review.
Providers and models
| Provider | Models |
|---|---|
| OpenAI | gpt-4.1-nano (default), gpt-4.1-mini, gpt-4.1, gpt-4o-mini, gpt-4o |
| Anthropic | claude-opus-5 (default), claude-sonnet-5, claude-haiku-4-5 |
The list is short on purpose. Every AI task here asks for structured output, and a model that cannot hold to a response schema passes a key check and then fails on the first real run — so the dropdown offers only models Meldoc runs.
Set up a provider
Prerequisites: Admin access to your workspace, and a plan that includes bringing your own key. Without it you see an upgrade prompt instead of the form.
- Open Settings → AI Provider.
- Select Setup AI Provider (or Change Provider to replace what is configured).
- In AI Provider, choose the provider.
- In API Secret, paste the key, then select Check key.
- In Model, choose one of the models the check found.
- Select Save Configuration.
The key is checked with the provider before anything is stored. The check asks the provider which models the key reaches and intersects that with the models Meldoc runs, so a typo, a revoked key or an organisation without access to a model is an error on the form rather than a silent failure in a background run later.
Until the check passes, the Model dropdown stays empty and Save Configuration stays disabled. Changing the provider or the key clears the result — a different key reaches a different set of models — so check again. A key that reaches none of the models Meldoc runs cannot be saved; pick a different key or provider.
Your API key is encrypted and never appears in API responses. The saved model is re-checked at save time, so a key that stops reaching its model is refused with the provider’s own words.
Change or remove a provider
Both start at Settings → AI Provider, in the Current Configuration card, which shows the provider, the model and Secret saved.
- Change Provider replaces the whole configuration: the current one is removed and you set a new provider, key and model.
- Remove returns the workspace to the shared quota, with standard plan limits. Features that need your own key stop working — the documentation review keeps posting, with the semantic half switched off.
What it costs
You pay your provider directly. Meldoc does not bill AI usage on your own key and does not meter it.
Where a run spends your key on your behalf, it says so: the documentation review’s pull request comment ends with the model, the tokens in and out for that run, an approximate cost at list prices where the model has one, and the total across every review run of that pull request.
AI Write and Ask AI still depend on your plan’s AI entitlements and daily request limits, shown under Settings → Plan & Billing → Limits & Usage. Connecting your own provider lifts the shared quota.
See Plans & Billing for what each plan includes.
What’s next?
Documentation Review — The review that runs on your own key.
Integration Tokens — Create tokens for CLI, MCP, and API access.