kind | string (one of) | | "chat" | chat, embedding | chat answers messages (for agents and applications); embedding turns text into vectors (for memories). Cannot be changed. |
model | object | | | | The model, as a vendor's catalog names it: the ModelProvider (this project's or the platform's) and the model id in it. Used with providers. |
model.provider | string | | | length 1 to 63 | Name of the ModelProvider, for example anthropic. |
model.id | string | | | length 1 to 128 | The model's id in that catalog, for example claude-opus-5-5. |
providers | list of InferenceProvider | | | up to 5 items | The InferenceProviders that serve the model, in the order to try them (routing and fallback). Empty for the built-in embedder. |
providers[].name | string | yes | | length 1 to 63 | Name of the InferenceProvider in this project. |
routing | object | | {} | | How the providers are tried: on a server error, a limit, an unreachable provider, a refused key or a model the provider no longer has, the next one is. |
routing.strategy | string (one of) | | "Ordered" | Ordered, Cost, Latency | Ordered tries them as listed; Cost the cheapest published price first; Latency the fastest measured first. |
provider | object | | {} | | Deprecated: one provider written into the endpoint. Use model and providers instead. A chat endpoint with none of these defaults to anthropic with model claude-opus-5-5; an embedding endpoint with none defaults to the built-in embedder. |
provider.type | string (one of) | | | anthropic, openai-compatible, builtin | anthropic (Messages API), openai-compatible (any provider with the OpenAI chat and embeddings API at baseUrl), or builtin (embedding only: a local hashing embedder that needs no key and costs nothing). |
provider.model | string | | | up to 128 characters | The provider's model id. Callers cannot choose another one. |
provider.baseUrl | string | | | up to 512 characters | For openai-compatible: the provider's API base address, https only, for example https://provider.example/v1. |
provider.credentialRef | object | | | | The Secret holding the provider's API key. Needed for anthropic and openai-compatible. Setting it needs the use right on the secret. |
provider.credentialRef.name | string | yes | | length 1 to 63 | Name of the Secret in this project. |
provider.credentialRef.key | string | | "api-key" | up to 63 characters | The key inside the secret that holds the API key. |
dimensions | integer | | | 1 to 8192 | Embedding only: length of the vectors. builtin defaults to 256. |
limits | object | | {} | | Quotas and rate limits of this endpoint. |
limits.requestsPerMinute | integer | | 60 | 1 to 6000 | Requests per minute for each calling identity. |
limits.inputTokensPerDay | integer | | 1000000 | 1 to 1000000000 | Input tokens per UTC day for the whole endpoint. |
limits.outputTokensPerDay | integer | | 200000 | 1 to 1000000000 | Output tokens per UTC day for the whole endpoint. |
limits.maxOutputTokens | integer | | 4096 | 1 to 64000 | The largest answer one request may ask for. |
logging | object | | {} | | Request and response logging, off by default. Needs monitoringRef; keys and authorization headers are never logged. |
logging.requests | boolean | | false | | Log request bodies. |
logging.responses | boolean | | false | | Log response bodies. |
monitoringRef | reference to MonitoringWorkspace | | | | Optional MonitoringWorkspace for request, token and latency metrics and, when enabled, logs. |