SMART_ROUTING_ENABLED | boolean | false | Enable the smart routing feature |
EMBEDDING_MODEL | string | text-embedding-3-small | Embedding model name used for vector search |
AZURE_OPENAI_EMBEDDING_MODEL | string | text-embedding-3-small | The actual OpenAI model name deployed in Azure (e.g. text-embedding-3-small). Azure deployment names are arbitrary identifiers, so this field tells Pcnaid MCP Hub which OpenAI model is behind the deployment, enabling correct token-limit enforcement and tokenizer selection. Only used when SMART_ROUTING_EMBEDDING_PROVIDER=azure_openai. |
EMBEDDING_MAX_TOKENS | number | By model | Maximum tokens for text truncation before generating embeddings. Overrides the per-model default. Useful to match the batch_size of a local inference server (e.g. LocalAI, LM Studio). When unset, the limit is resolved automatically: text-embedding-* → 8191, bge-m3 → 8192, gemini-embedding-001 → 2048, other BGE models → 512, unknown models → 512. |