Models Component¶

The Models component turns provider, model, and embedding definitions scattered across config into a unified registry that the rest of the system can actually use. Agent, RAG, and Tools are all built on top of it.

1. What problem it solves¶

Different model providers vary greatly in API style, authentication, and capabilities. If every business component integrated with upstream providers directly, the system would become hard to maintain quickly. Models exists to converge those differences into one entry point.

2. What capability it provides¶

initialize the Genkit Registry
register chat models
register embedding models
set the default model
convert provider config into actual plugins

3. Provider shapes currently supported¶

From the config and implementation, the project is currently organized mainly around OpenAI-compatible endpoints and a few specific plugins:

DashScope
Gemini
SiliconFlow
Pinecone plugin capability

Whether they are actually active depends on whether API keys are present.

4. Initialization process¶

flowchart TD A["Read provider config"] --> B["Filter providers without API keys"] B --> C["Create plugins"] C --> D["genkit.Init(...)"] D --> E["Register default model"] E --> F["Register models and embeddings"] F --> G["Store Genkit Registry in Runtime"]

5. Example config¶

type: models
spec:
  default_model: "dashscope/qwen-max"
  default_embedding: "dashscope/text-embedding-v4"
  providers:
    dashscope:
      api_key: "${DASHSCOPE_API_KEY}"
      base_url: "https://dashscope.aliyuncs.com/compatible-mode/v1"

Key fields:

default_model: the model used by Agent by default
default_embedding: the embedding model used by RAG and related capabilities
providers.*.models: chat, code, or multimodal model list
providers.*.embedders: embedding model list

6. Why it is a system-level component¶

Models is not only used internally. It provides the unified Genkit Registry for the whole project:

Agent uses it to define prompts and execute flows
Tools uses it to register tools
RAG uses it for embedding and retrieval integration

Without Models, many later components may exist in code but cannot actually function.

7. Current limitations and caveats¶

genkit.Init() is currently only suitable to run once, so Models behaves more like a global singleton capability.
Provider extension is still relatively centralized and usually needs new plugin creation logic inside the component.
Listing a model in config does not guarantee that model will work. Real availability still depends on provider API keys and upstream support.

8. What to check when troubleshooting Models¶

whether the default model name is correct, usually provider/model
whether the API key was really injected
whether the provider base_url is reachable
whether the model was registered into the Genkit Registry
whether the upstream model type matches intended usage, such as chat, code, or multimodal