SearchSpace API

Grab a key from the console and send your first query.

curl https://q.searchspace.io/v1/search \
  -H "authorization: Bearer $SEARCHSPACE_API_KEY" \
  -H "content-type: application/json" \
  -d '{ "query": "how do vector databases work", "top_k": 5 }'

Every request needs an API key. Create one in the console; it's shown only once, so store it securely and pass it as a Bearer token (or x-api-key). Keep keys server-side — never ship them in client code.

# Pass your key with either header — both are accepted.
authorization: Bearer sk_live_xxxxxxxxxxxxxxxxxxxxxxxx
x-api-key: sk_live_xxxxxxxxxxxxxxxxxxxxxxxx

/v1/search is the primary endpoint. The Exa-compatible routes accept Exa's request/response shape, so an existing Exa SDK works unchanged by pointing it at the base URL above.

A search is a JSON body with a query and any of the optional fields below.

The filters object narrows results by metadata. Within a field, values are OR'd; across fields they are AND'd. Dates are Unix seconds.

200 OK returns a results array and a latency_ms timing. Each result carries the fields below; text, highlights and highlight_scores appear only when requested via contents.

{
  "results": [
    {
      "doc_id": "9f2c4e1a7b8d...",
      "url": "https://example.com/vector-databases",
      "title": "How vector databases work",
      "snippet": "...approximate nearest-neighbor search over embeddings...",
      "score": 0.94,
      "crawl_date": "2026-06-11",
      "domain": "example.com",
      "language": "en"
    }
  ],
  "latency_ms": 38
}

The index includes GitHub source code, segmented by language. Restrict to all code with filters.category: ["github_code"], to a language with code_rust / code_python, and gate on repo popularity with filters.min_stars / filters.max_stars. A file carries both github_code and its language category (categories are multi-valued), so the two compose.

curl https://q.searchspace.io/v1/search \
  -H "authorization: Bearer $SEARCHSPACE_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "query": "lock-free ring buffer",
    "top_k": 10,
    "filters": {
      "category": ["code_rust"],
      "min_stars": 1000
    }
  }'

There's no separate document-fetch call. Ask for page text or query-relevant highlights right in the search request with the contents object, and they come back inline on every result.

curl https://q.searchspace.io/v1/search \
  -H "authorization: Bearer $SEARCHSPACE_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "query": "how do vector databases work",
    "top_k": 5,
    "contents": {
      "text": { "max_characters": 2000 },
      "highlights": { "num_sentences": 3 }
    }
  }'

Limits are enforced per API key. Exceeding them returns 429; back off briefly and retry.

Errors use standard HTTP status codes and return a JSON body with an error message. 5xx responses are safe to retry with backoff.