ConareDB quickstart

Get a key

Provision a sk-conare-... key from the ConareDB console. Keys are scoped to your namespace prefix with read/write permissions, shown once, and sent as a Bearer token:

export CONAREDB_API_KEY='sk-conare-...'

Write rows

Namespaces are created on the first write — no schema, no ceremony. Each row is a node with an external id, properties, and an optional vector. The vector dimension is fixed by the first vectored write and validated on every subsequent one.

curl -s https://db.conare.ai/v1/namespaces/acme-docs/query \
  -H "Authorization: Bearer $CONAREDB_API_KEY" \
  -d '{
    "request_type": "write",
    "query": {
      "queries": [{"Query": {"name": "w", "condition": null, "steps": [
        {"UpsertN": {"id": "chunk-123", "label": "chunk",
          "props": {"text": "Refund policy: 30 days, no questions.", "created_at_ms": 1720000000000, "project": "docs"},
          "vector": [0.0123, -0.0456]}}
      ]}}],
      "returns": ["w"]
    }
  }'

UpsertN is a full-row replace keyed on id — write the same id again and the old row (and its vector) is replaced, not merged.

One request runs up to three ranked branches — vector ANN, BM25 keyword, and chronological — over the same filtered row set:

curl -s https://db.conare.ai/v1/namespaces/acme-docs/search \
  -H "Authorization: Bearer $CONAREDB_API_KEY" \
  -d '{
    "label": "chunk",
    "filter": ["project", "Eq", "docs"],
    "include_props": ["text"],
    "branches": {
      "vector":  {"embedding": [0.0119, -0.0442], "top_k": 10},
      "keyword": {"query": "refund policy", "top_k": 10}
    }
  }'

{
  "vector":  {"hits": [{"id": "chunk-123", "score": 0.98, "props": {"text": "..."}}], "server_us": 412},
  "keyword": {"hits": [{"id": "chunk-123", "score": 14.2, "props": {"text": "..."}}], "mode": "and", "server_us": 200}
}

Branches return independent ranked lists so you control fusion (RRF or your own) client-side — the same architecture Conare’s own memory plane uses.

Need vectors? POST /api/v1/embeddings on the Embeddings API returns unit-normalized 1024-dim vectors. Use input_type: "document" for rows you store and input_type: "query" for the vector branch that searches them.

Next: the full search reference, writes & bulk ingest, or namespaces & keys.

Get started

Reference

Guides

ConareDB quickstart