Research
In the product: open codefundi.app/research (or Search with Research mode). API: same endpoint as search, POST /v2/search with chat: true. Response is application/x-ndjson, not a single JSON object. There is no separate /v2/research route.
Research runs a normal codebase search, then streams an AI analysis grounded in the top hits. Ideal when you need both retrieval and synthesis (for example: how auth flows through services, or where billing is validated before persistence).
How it behaves
Search executes
The service runs search (semantic or grep) with your query, scopes, and filters.
Evidence-backed answer
Hits appear first, then the model answers using that evidence. Context size follows your plan (Credits and tiers).
Stream completes
The product consumes an NDJSON stream: search, then chunk text, then done (with model and context_files) or error.
How to use it in the web app
- Open Research (
/research), or switch the search box to Research mode on/search. - Select repos and filters as you would for search.
- Submit a question. Review ranked evidence, then the streamed answer.
Credits: search base charge plus LLM usage for the streamed answer. Expect 402 when balance is insufficient.
Same capability via API
POST /v2/search with "chat": true, Accept: application/x-ndjson, and optional model. History logs the request under category research. Details: V2 Search.