Package-level declarations
Types
Hybrid compressor that uses different strategies based on text length.
Unconditionally mints a fresh NewEntity for every suggestion.
Built-in prompt strategies for LlmCandidateBakeoff.
Strategy for building prompts used in LLM candidate bakeoff.
Structured response from the bakeoff LLM.
Selects the best match from multiple candidates.
A searcher that finds candidate entities for a suggested entity.
Entity resolver that delegates to multiple resolvers in order.
Compresses source text to extract only the portions relevant to entity resolution.
Entity resolver that walks a chain of CandidateSearchers from cheapest to most expensive, stopping as soon as one returns a confident match. If no searcher is confident, the accumulated candidates go to an optional LLM bakeoff; if that also finds nothing, a new entity is minted (or vetoed if the schema forbids creation).
Entity resolver that remembers entities it's been asked to resolve and tries to reuse them. Useful for deduplicating entities within a single session.
Entity resolver decorator that checks a list of known entities before delegating to another resolver.
The outcome of a single resolution attempt, including which level resolved it and how many candidates were considered.
Uses an LLM to select the best match from multiple candidates.
No-op compressor that returns text unchanged.
Which strategy successfully resolved an entity. Lower levels are faster and cheaper; higher levels pull in the LLM.
Result of a candidate search operation.
Extracts context by finding sentences containing the entity.
Structured response for single candidate verification.
Extracts a window of text around each mention of the entity.