Dify

Dify supports adding a custom OpenAI-API-compatible model provider, which is how it talks to Modelship.

1. In Dify, go to Settings → Model Provider and add an OpenAI-API-compatible provider (Dify lists this as a distinct provider type from "OpenAI").
2. Set the API Base to http://<modelship-host>:8000/v1.
3. Set the API Key to a real key if MSHIP_API_KEYS is configured, or any non-empty placeholder if not.
4. Add a model entry for each model name in your models.yaml (Dify needs the model registered explicitly per provider — it doesn't auto-discover via GET /v1/models), matching its type (chat, embeddings, etc.) to the corresponding Modelship deployment.

Once registered, use that model like any other inside Dify's chatflow/agent builder. Streaming, embeddings (for Dify's knowledge-base retrieval), and tool calling all route through the same /v1 endpoints Modelship already exposes.