Docs

Create an assistant and run it against the calling API key’s usage.

Assistants

Assistants are first-class API objects. Create one, then call POST /v1/assistants/:id/chat. That run uses the assistant’s modelId and systemPrompt, optional live web search, and debits the Bearer key the same way POST /v1/chat/completions does.

GET    /v1/assistants
POST   /v1/assistants
GET    /v1/assistants/:id
PATCH  /v1/assistants/:id
DELETE /v1/assistants/:id
POST   /v1/assistants/:id/chat

Create

bash
curl http://localhost:3001/v1/assistants \
  -H "Authorization: Bearer $OPENDOOR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Docs bot",
    "modelId": "opendoor/auto",
    "systemPrompt": "Answer from the OpenDoor docs. Be short.",
    "webSearchEnabled": false
  }'

webSearchEnabled: true requires the Web Search add-on (402 otherwise).

Run (usage-billed)

bash
curl http://localhost:3001/v1/assistants/$ASSISTANT_ID/chat \
  -H "Authorization: Bearer $OPENDOOR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [{"role": "user", "content": "What is an API key spend cap?"}],
    "tools": [],
    "temperature": 0.2
  }'

The response is an OpenAI-shaped chat completion plus assistant_id and assistant_slug. Tokens and cost land on GET /v1/usage and GET /v1/requests for this org.

You can still pass tools / tool_choice / response_format — they go to the same provider path as chat completions.

SDK

ts
const a = await client.assistants.create({
  name: "Docs bot",
  modelId: "opendoor/auto",
  systemPrompt: "Be brief.",
});
const out = await client.assistants.chat(a.id, {
  messages: [{ role: "user", content: "Hello" }],
});

Related

Previous

Usage

Next

Workflows