Search Lab
Compare two retrieval configurations on hand-labelled questions about real code.
demo/minirouter: The bundled demo fixture: a small Python WSGI router written for RepoKnit's previews. 4 files.
Retrieval runs in your browser over source read at build time; nothing is sent to a paid provider.
Questions and expected ranges were written by hand by the RepoKnit author against the source at the time of writing, and an automated test checks that every expected range's anchor still resolves to exactly one line.
No query selected. Choose a benchmark question or type a custom query.
Configurations
Configuration A
Method
- Semantic: Semantic retrieval needs an embeddings backend, which isn't built yet.
- Hybrid: Hybrid retrieval blends BM25 with semantic retrieval, which needs an embeddings backend that isn't built yet.
Chunking
Indexing
1 to 20
Only chunks whose path starts with this
Configuration B
Method
- Semantic: Semantic retrieval needs an embeddings backend, which isn't built yet.
- Hybrid: Hybrid retrieval blends BM25 with semantic retrieval, which needs an embeddings backend that isn't built yet.
Chunking
0 to 19
Indexing
1 to 20
Only chunks whose path starts with this
Run a query to compare A and B side by side, or run the benchmark to score both on every labelled question.
Run history
Not saved: runs are kept until you leave the page.
No runs yet.