Rerank Text (Ollama)

@ollama1 VersionQuery / AI

LLM-as-reranker over a local Ollama runtime. Sends the query and documents from a RerankInput to a chat model, parses its JSON relevance scores, sorts descending, optionally truncates, and emits [RerankResult] with each document, score and original index.

How it fits

Typical backends

Rerank retrieved passages and serve the ordered results as JSON.

Rerank, serialise the ordered list, then push it over HTTP.

Caveats

FallbackWhen the model reply cannot be parsed as JSON the original document order is preserved with zero scores rather than failing; markdown code fences and a bare top-level array are both tolerated when parsing succeeds.
AccuracyRanking quality and stability depend on the chosen chat model; keep temperature at zero for deterministic, repeatable orderings.
Parameter interactiontop_k zero returns every document, appending any the model did not score at the tail with a zero score to preserve coverage; a positive value truncates after sorting and drops unscored documents entirely.
I/O contractScores the model reports for an index outside the candidate range are discarded, and each emitted RerankResult keeps the document's original position in the RerankInput list as its index.
CompatibilityThis component requires the Ollama runtime it depends on to be reachable; the chat model is pulled to the host once on first use before any message is served.
State lifetimeThe model selection and all configuration are captured once on first invocation; changing them requires a redeploy.

Versions

76719beclatestdefault
Automated release
8/2/2026