Create a chat completion

POST /projects/:projectid/chat/completions

Generate a response using an Intellectible model tier.

Authenticates a project API key, checks that it may use the requested tier, then routes small and medium to Intellectible's in-house model infrastructure or large to the external large-model provider.

Authentication

• Project API key: The credential for everything external. Created in Project Settings, bound to one project, and limited to the scopes it was granted. Format: proj.<keyId>.<secret>.
• Required permission: llm.complete on :model.

Path parameters

Name	Type	Required	Default	Description
projectid	string	Yes	-	The project id.

Request body

Content type: application/json

Name	Type	Required	Default	Description
model	string	Yes	-	One of small, medium, or large. These are stable capability tiers rather than model IDs.
messages	array	Yes	-	OpenAI-style chat messages.
stream	boolean	No	false	Request a streamed response.

Example

{
  "model": "medium",
  "messages": [
    {
      "role": "user",
      "content": "Hello"
    }
  ],
  "stream": true
}

Example request

curl -X POST \
  -H "Authorization: proj.KEY_ID.SECRET" \
  -H "Content-Type: application/json" \
  -d '{"model":"medium","messages":[{"role":"user","content":"Hello"}],"stream":true}' \
  {{BASE_URL}}/projects/{{PROJECT_ID}}/chat/completions

Responses

200

The completed OpenAI-style chat completion when stream is false.

{
  "id": "chatcmpl_abc123",
  "object": "chat.completion",
  "model": "medium",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Hello!"
      },
      "finish_reason": "stop"
    }
  ]
}

200 — stream

An SSE stream of OpenAI-style completion chunks followed by data: [DONE] when stream is true.

400

The tier is unknown or the request is malformed.

401

Missing, invalid, or disabled project API key.

403

The key is not permitted to use this tier.

502

The queue service, in-house worker, or external provider could not start, stream, or complete the request.