Microsoft Foundry

OpenDocBot connects to Microsoft Foundry through its v1 API, the OpenAI-compatible surface Microsoft recommends.

Setup:

1. Preset: Microsoft Foundry
2. In the Foundry portal, deploy a model and note its deployment name
3. Copy the resource Endpoint and one of its Keys
4. Paste the endpoint and key, then type the deployment name into Model
5. Test Connection, Apply

Notes:

• Endpoint. Paste the endpoint from the Foundry portal. Any resource URL works: the add-in normalizes it to the canonical v1 base https://<resource>.services.ai.azure.com/openai/v1. That includes the resource root, /models, /models/chat/completions, or a full /openai/v1/chat/completions?api-version=... URL. The v1 API is also served on https://<resource>.openai.azure.com/openai/v1 (classic Azure OpenAI resources); use whatever your portal shows.
• Model = deployment name. The Model field is your deployment name (or the model id for serverless endpoints), not the base model's public name. OpenDocBot does not query the Foundry catalogue; enter the name exactly as it appears in the portal, for example gpt-oss-120b or DeepSeek-V4-Flash.
• No api-version. The v1 API does not use dated API versions.
• Legacy Model Inference API is not supported. Endpoints of the form /models/chat/completions?api-version=... are not used by this preset; the v1 base above is.
• Reasoning and prompt caching work where the underlying model supports them.
• Browser blocked (CORS). If the taskpane cannot call the endpoint directly, enable Proxy API requests through this server (self-hosted / dev deployments) to route the request through the same-origin /proxy/ endpoint.

The API key is stored in browser localStorage like every other provider key and is sent only to your Foundry endpoint. Treat it as a secret and rotate keys you no longer use.