14 integrations support setting a custom URL/endpoint.
| Integration | Field | Required? | Notes |
|---|---|---|---|
openai |
url |
optional | Enables any OpenAI-compatible server |
cohere |
url |
optional | Shared across all Cohere task types |
hugging_face |
url |
required | Points to your HuggingFace Inference Endpoint |
azure_ai_studio |
target |
required | Full Azure ML deployment URL; path suffixes appended internally |
deepseek |
url |
optional | Falls back to hardcoded DeepSeek endpoint |
llama |
url |
required | No default; designed for self-hosted endpoints |
nvidia |
url |
optional | Falls back to per-task NVIDIA NIM defaults |
openshift_ai |
url |
required | Self-hosted OpenShift AI Model Serving |
watsonxai |
url |
required | IBM Cloud regional endpoint |
google_vertex_ai |
url + streaming_url |
optional | Chat completion only, and only when provider ≠ GOOGLE (Model Garden third-party endpoints) |
groq |
url |
optional | Falls back to default Groq endpoint |
fireworks_ai |
url |
optional | Separate defaults per task type |
custom |
url |
required | The whole point of the integration |
alibabacloud-ai-search |
host + http_schema |
host required |
Constructs URL from host + schema |
| Integration | Reason |
|---|---|
anthropic |
Hardcoded endpoint; no override |
azure_openai |
URL constructed from resource_name + deployment_id + api_version |
amazon_bedrock |
URL determined from region + provider + model |
google_ai_studio |
Hardcoded endpoint; no override |
google_vertex_ai (embeddings + rerank) |
URL constructed from project_id + location + model_id |
mistral |
Hardcoded endpoint; no override |
jinaai |
url field was present but removed in a refactor |
voyageai |
Hardcoded endpoint; no override |
contextualai |
url field was removed |
ai21 |
Hardcoded endpoint; no override |
amazon_sagemaker |
AWS-managed routing via endpoint_name, region, api |
elasticsearch |
Routes to local ML inference nodes |
elastic |
Node-level config only via xpack.inference.elastic.url in elasticsearch.yml |