5. Chat completions - co-mind.ai

Chat completions are stateless — the platform doesn’t remember prior turns. You send the full conversation in messages[] on every request; the server returns a single completion. Same shape as the OpenAI Chat Completions API, so existing tooling works.

Send a chat request

curl -X POST {BASE_URL}/v1/chat/completions \
  -H "Authorization: Bearer $PAT" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "'$CHAT_MODEL'",
    "messages": [
      {"role": "user", "content": "In one sentence, what is retrieval-augmented generation?"}
    ]
  }'

{
  "id": "chatcmpl-xyz",
  "object": "chat.completion",
  "created": 1756820000,
  "model": "llama3.2:3b",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Retrieval-augmented generation combines a language model with a search step over external documents so answers can cite fresh, domain-specific sources."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": { "prompt_tokens": 18, "completion_tokens": 34, "total_tokens": 52 }
}

Follow-up turns

For a multi-turn conversation, append each assistant response to messages[] and send the whole thing again:

{
  "model": "llama3.2:3b",
  "messages": [
    {"role": "user", "content": "In one sentence, what is retrieval-augmented generation?"},
    {"role": "assistant", "content": "Retrieval-augmented generation combines..."},
    {"role": "user", "content": "Give me a concrete example."}
  ]
}

Streaming

Add "stream": true to receive Server-Sent Events instead of one JSON response. Frames arrive as data: {...} lines and end with data: [DONE].

For chat grounded in your documents rather than raw model knowledge, use the knowledge base chat endpoint in the next section.

Next: Knowledge bases →