Configuring AI Providers

Configure which AI providers and models are available for agent runs, and enforce an organisation-wide model allow-list.

Supported providers

OpenAI: every model your OpenAI account can call
Anthropic: every Claude model your API key can call
GitHub Models: the multi-vendor catalogue on the GitHub Models marketplace, billed against your GitHub token
Azure OpenAI: your own Azure deployments — requires an endpoint URL, and models are chosen by deployment name
Custom: Any OpenAI-compatible endpoint

How the model list is built

Productivity Workbench does not ship a fixed list of models — there is nothing to go stale. Once you have applied and tested an API key, click Fetch Models List (or use Apply, Test & Fetch Models, which does all three in one step) and Model Config calls the provider’s own models API and shows exactly what your account can use today. New models appear as soon as your provider releases them; retired ones drop off. Azure OpenAI returns your deployment names rather than model IDs, and Custom providers are always entered by hand.

Adding a provider

Go to Model Config → AI Providers. Click Add [Provider]. Enter your API key and click Apply — the key is stored as an opaque reference and is never displayed again. Set default model and temperature (0.0–0.3 for code; 0.5–0.8 for creative tasks). Test the connection.

Model Allow-List

Configure approved models in the Model Allow-List. Agents will not use models outside this list regardless of what providers offer. Leave empty to permit all models from configured providers.

Cost considerations

Cost varies by an order of magnitude across a single provider’s range. The flagship reasoning models are the most capable and by far the most expensive; the smaller, faster tiers are often sufficient for classification, extraction and summarisation. Model names and prices change often, so check your provider’s current pricing page rather than any list here — then use per-run cost data in AI Effectiveness to see real spend by model, and Max Cost Per Run to cap it.

AI cost and Workbench credits

A studio shows two different measures, and they are never added together.

AI cost — what your own AI provider charges for the calls your agents make, in dollars. It is estimated from the tokens each provider reported, and it is billed to you by your provider, on your own key. See it on AI Spend and, rolled up the work, on Cost by Intent.

Workbench credits — what the studio used of the Workbench itself: compute time, reading and writing its records, requests and data sent. Credits are a measure of use, not money. They are included in your plan today and nothing is charged for them. See them on Workbench Credits, beside the AI cost on Cost by Intent, and in the header of every work package.

Opening a page or saving a record uses a small fraction of a credit. An agent run or an Auto Pilot run uses more, because it works for longer and writes more. Every figure is an estimate.

All AI Product Studio documentation

Build v0.3.2 · deployed 2026-10-19T23:52:31Z