Developer access · Field notes
Read cached research before starting another search
A small client workflow that avoids unnecessary provider work and handles interrupted streams honestly.
Begin with the exact query
The startidea public API exposes cached research without triggering a provider call. Read /api/index to find a workflow, then GET /api/search with its query. A cache miss is a 200 response whose search field is null. This is deliberate: an empty cache is not a transport failure and does not prove that no companies serve the market.
Choose the response mode deliberately
An authorized POST can use normal JSON or application/x-ndjson. JSON returns a single result and checks up to the first three companies. The streaming mode publishes a small initial set, adds matches, and checks the returned companies in batches. Each results event is a complete snapshot. Replace your current snapshot instead of appending the same companies again.
A newline is a frame, not a network chunk
An HTTP chunk can end halfway through a JSON object or a multibyte character. Decode incrementally, buffer until a newline, then parse each event. A complete event marks success; an error event or premature EOF means the process did not report successful completion to your client. Keep the last usable results visible and label them incomplete.
Do not retry a paid POST blindly
A connection can disappear after work has started. The API has no durable replay key, so repeating an ambiguous request may perform provider work again. Read the cached query and compare its source timestamps first. A rate-limit rejection includes Retry-After; that is different from an unknown outcome after a timeout.
Sources & scope
Prepared October 13, 2026 with AI assistance. Vendor descriptions are attributed claims. Worksheets are proposed methods; numerical examples are illustrative. No hands-on vendor testing is represented.