Testing
upload → load the models → ask · Context returns the passages that fit the budget, Agent writes the answer models: idle◷ wasm engine · loads on first use
1 Add files — or continue with sources you've already loaded
2 Pick an answer model — the rest load automatically
Answer modelloading…
Rerankerloading…
Semantic embedderloading…
Tokenizerloading…
Configuration
passages below this relevance are dropped
kept passages ≥ this expand their neighbours; normally BELOW the keep threshold
Context returns what fits in this budget
Agent mode only
3 Ask a question about what you loaded
Context · the passages that survive reranking and fit the token budget — retrieval only, no model
Response
Ask a question and press Run.
Pipeline stages — click a step to inspect it