Skip to main content
Custom endpoints let you route inference requests to any service that implements a supported API specification. This covers self-hosted models, enterprise proxy servers, and cloud platforms like Azure OpenAI or AWS Bedrock.

​
Creating a custom endpoint

1

Open Models & Providers

Navigate to Settings → Agent → Models & Providers and scroll to the Custom Endpoints section.
2

Add a new endpoint

Click Add endpoint and fill in the required fields.
3

Configure the endpoint

Provide:
  • Name — A display name for this endpoint
  • API spec — Which API protocol the endpoint implements
  • Base URL — The full base URL of the endpoint
  • API key (optional) — Authentication key for the endpoint

​
Supported API specifications

SpecDescriptionExample services
anthropicAnthropic Messages APISelf-hosted Claude, any Anthropic-compatible proxy
openai-chat-completionsOpenAI Chat CompletionsvLLM, Ollama, LiteLLM, Together AI
openai-responsesOpenAI Responses APIDirect OpenAI-compatible responses endpoints
googleGoogle Generative AISelf-hosted Gemini-compatible services
azureAzure OpenAIAzure OpenAI Service
amazon-bedrockAWS BedrockAmazon Bedrock (requires region + secret key)
google-vertexGoogle Vertex AIVertex AI (requires project ID + location + credentials)

​
Cloud provider configuration

​
Azure OpenAI

Azure endpoints require additional fields:
  • Resource name — Your Azure OpenAI resource name
  • API version — The API version string (e.g. 2024-02-01)

​
Amazon Bedrock

Bedrock endpoints require:
  • Region — AWS region (e.g. us-east-1)
  • API key — AWS Access Key ID
  • Secret key — AWS Secret Access Key

​
Google Vertex AI

Vertex endpoints require:
  • Project ID — Your GCP project ID
  • Location — GCP region (e.g. us-central1)
  • Google credentials — Service account JSON credentials

​
Model ID mapping

Some custom endpoints use different model identifiers than the official APIs. Use the Model ID Mapping field to remap built-in model IDs to the IDs your endpoint expects. For example, if your Ollama instance names Claude as claude-v2: 
{
  "claude-sonnet-4-6": "claude-v2"
}

​
Using a custom endpoint

After creating the endpoint, go to the provider configuration for any provider and select Custom endpoint mode. Choose your endpoint from the dropdown. All models for that provider will route through the custom endpoint.