Model features & the /_chat facade #

Every model in the AI-services registry can now declare what it is for and what it accepts — features (embedding or generation) and inputs (text or image). The two axes are orthogonal: “multimodal” is not a category of its own but an inputs list beyond text.

node:
  embedding:
    endpoints:
      - name: zai
        url: https://api.z.ai/api/coding/paas/v4/embeddings
        chat_url: https://api.z.ai/api/coding/paas/v4/chat/completions
        api_key: sk-…
        models:
          - embedding-3                                   # embedding + text (the defaults)
          - {id: GLM-4.6V, dims: 3072}                    # embedding, dims declared
          - {id: glm-4-flash, features: [generation]}     # text generation
          - {id: glm-4v, features: [generation], inputs: [text, image]}   # vision LLM
          - {id: clip-v2, features: [embedding], inputs: [text, image]}   # multimodal embedder

Omitted fields keep the historical defaults, so every existing plain-string listing still means embedding + text — nothing to migrate.

The features are enforced where they matter, not just displayed:

• Semantic bindings (semantic_text, semantic multi-fields, source-mapped dense_vector) reject a generation-only listing at mapping time (“not an embedding model”) instead of failing at write time, and the console’s model picker lists embedding models only.
• Generation gets its own facade, the sibling of /_embedding:

POST /_chat
{
  "model": "glm-4-flash",
  "messages": [{"role": "user", "content": "hello"}],
  "temperature": 0.2
}

The body is the standard chat-completions shape (plus an optional service disambiguator that pizza strips); it routes to the service listing the model with the generation feature, and the provider’s answer — choices, usage, provider extras — passes through verbatim. The management API follows suit: services carry a chat_url, and the connectivity probe tests a model the way it serves (embedding models answer their dimensionality, generation models answer a reply excerpt).