AI providers
Open AI & models → AI Providers (/admin/settings/ai/providers) to configure vendor credentials shared by server evaluations, Insights, session titles, outcome summaries, and work-item agents. Choose the provider and model on the page for each feature; embedding credentials have their own Embeddings page.
Add a provider
Section titled “Add a provider”- Enter its API Key, and an optional Base URL where supported.
- Set Max Parallel Judges to bound simultaneous judge calls for that vendor.
- Click Save.
- Open Evals, Insights or Summaries, or Work Items to configure the feature that should use it.
An already-configured secret stays unchanged when its field is blank. To remove it, select Clear on save and save. Deployment-supplied values may be read-only.
Vendor settings
Section titled “Vendor settings”The table uses the configuration prefix for each vendor. For example, Anthropic’s key is Evals:ServerRunner:Anthropic:ApiKey.
| Vendor | Configuration prefix | Credentials and endpoint |
|---|---|---|
| Anthropic | Evals:ServerRunner:Anthropic | ApiKey |
| OpenAI | Evals:ServerRunner:OpenAI | ApiKey |
| Cohere | Evals:ServerRunner:Cohere | ApiKey; optional BaseUrl for an on-premises deployment |
| OpenRouter | Evals:ServerRunner:OpenRouter | ApiKey; BaseUrl defaults at runtime to https://openrouter.ai/api/v1 |
Each of these vendors also has MaxParallelJudges, from 1–16, default 4. Judge model fields live under Evals → Eval judge and use Evals:ServerRunner:<Vendor>:Model. OpenRouter models use the full provider/model identifier.
For hosted Cohere, supply an API key. For an on-premises deployment, set the API base URL, such as https://cohere.internal, without /v2/chat; the key is optional for deployments that do not require one.
Kurrent AI
Section titled “Kurrent AI”Kurrent AI uses credentials and endpoints managed by Kurrent. Commercial tenants do not edit its token, judge model, or concurrency here; the Kurrent AI configuration section is only exposed where those deployment settings are editable.
Managed titles, summaries, embeddings, and evaluations are available on every plan, including Free. Insights and Work Items require Team or Enterprise. See Plans & seats for the full distinction between feature access and provider availability.
Model pricing
Section titled “Model pricing”The Model pricing panel reports the source, number of priced models, last successful sync, and any last error. Rates sync automatically from models.dev. A warning appears when no sync has succeeded or the last success was more than 48 hours ago. Recorded costs remain estimates.