Build with the AI APIs
This section is for developers who use the platform's AI from their own applications, scripts and tools. You reach everything over HTTP with an API key, and you can keep using the OpenAI or Anthropic client you already know.
With the AI APIs you can:
- call models through one endpoint, the Model Gateway, whether they run in your cluster or at an external provider
- give a model tools, such as search over your own documents or your own MCP servers, and let the gateway run the tool calls
- upload documents to a RAG endpoint and retrieve context from them
- use the platform's models in coding agents and in VS Code
Every request carries an API key as a bearer token: Authorization: Bearer cm_api_…. The key's grants decide which
models and RAG endpoints it can reach. The Model Gateway's address is the API URL at the bottom of the portal
sidebar. Each RAG endpoint has its own URL, shown on its page in the portal.
Start with the page that fits what you want to do:
- Quickstart: your first Model Gateway call: in about 10 minutes you create an API key and call a
model with
curlor the OpenAI Python client. - Create and manage API keys: personal and service-account keys, grants, rotation and usage.
- Call models through the Model Gateway: the endpoints, model IDs and aliases, and what each model supports.
- Responses API: the recommended API for new integrations, with streaming and stored conversations.
- Messages API: call models with the Anthropic Messages request shape.
- Use Anthropic models: connect Anthropic models and use them through the Model Gateway.
- Tools and preset agents: let a model search your documents, call MCP tools, or run as a preset agent.
- RAG endpoint API: upload documents to a RAG endpoint and retrieve context over its API.
- Configure coding agents and Configure VS Code agent mode: use the platform's models in your coding tools.