ContextX — find the line a passage came from

Query

Try

Retrieved

What runs here, and what doesn't

Here: the ranking from the repository's offline walkthrough adapter, ported to JavaScript. Query and notes are split into lowercase words; each note is ranked by how many distinct query words it contains. Every word counts, including “the”. Nothing is sent anywhere.

In ContextX: uploaded text is chunked (1,000 characters, 100 overlap), embedded by a Google model, stored in Pinecone and retrieved by vector similarity; answers come from an LLM. None of that runs on this page, so paraphrases like “meeting” for “planning” won’t match here.

Differences: this page hides notes with no shared words (the fixture returns its top results regardless) and quotes the best-matching line, where the service's reference quotes the chunk's first 200 characters.

Sources

Synthetic notes. Matched words are marked; a citation link jumps to its line.