4. Compute embeddings - co-mind.ai

Embeddings turn text into a fixed-length numeric vector so you can do similarity search, clustering, or reranking outside the platform. If you want retrieval handled for you, skip ahead — Knowledge Bases manages embedding, indexing, and retrieval end-to-end.

Send a request

OpenAI-compatible shape. Use one of the embedding model ids from the previous step.

curl -X POST {BASE_URL}/v1/embeddings \
  -H "Authorization: Bearer $PAT" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "'$EMBED_MODEL'",
    "input": "co-mind.ai is a private AI platform"
  }'

{
  "object": "list",
  "data": [
    {
      "object": "embedding",
      "index": 0,
      "embedding": [0.012, -0.087, 0.034, "..."]
    }
  ],
  "model": "nomic-embed-text",
  "usage": { "prompt_tokens": 8, "total_tokens": 8 }
}

Vector length depends on the model — nomic-embed-text returns 768 floats, other models are different. The data[0].embedding array is what you feed into your vector store.

Batch multiple inputs in one call by sending "input": ["text one", "text two", "text three"]. Each input gets its own entry in data[], matched by index.

Next: Chat completions →