Try it now
A working sandbox. No sign-up, no project. Sample data only.
- Four modes: Portraits, Ensemble, Timeline Chronicle, Thematic Assembly.
- Optional sections: Serendipities, Facts & Timelines, Editor's Note.
- Only what the speaker said on tape, with citations kept.
- Export to .docx.
Try a preset query or type your own.
How it works
Three steps from raw material to result.
Each transcript is broken into passages and indexed for meaning as soon as it lands. The project archive grows automatically with every new file.
Type the idea, not the exact words. The search ranks every passage in your archive by how closely it matches what you're looking for.
The numbers stopped making sense. That was the moment we knew.
Every result carries the file, timecode, and speaker. Click any hit to open the player at that line.
Quotes and capitals
Quotes hold a phrase together. Capitals keep an abbreviation out of ordinary words.
Frequently asked questions
How is this different from keyword search?
Keyword search needs the exact word. Semantic search ranks lines by meaning. A query like "when the numbers stopped making sense" finds "the math wasn't adding up" too.
What's under the hood?
Each transcript passage is encoded into a meaning-based index. Your query is encoded the same way and ranked by how closely it matches, with a small reranking pass on the top results.
Can I search across all my projects at once?
No. Semantic search is scoped to a single project to keep results focused and to respect collaborator boundaries. Use the project switcher to move between archives.
Do I need to re-index when I add a file?
No. New transcripts are indexed automatically in the background as soon as they land. You'll see a small progress indicator on the project page until indexing completes.
Is there a chat interface on top of search?
Yes. You can ask questions about your project and get answers grounded in cited transcript passages, with every citation showing speaker, file, and timecode.
How do I search for a short abbreviation?
Type it in capitals. Searching "EV" in lower case matches the inside of every, even, seven, level and eleven, and the handful of real hits drown. In capitals it is matched at word boundaries and through the spellings a transcriber reaches for, so E.V. and EV both come back and "eleven" does not.
How do I search for an exact phrase?
Put it in quotation marks. Unquoted, the words are hunted separately, so a passage saying "expected" in one sentence and "value" in the last scores as highly as one that actually says "expected value." Quotes hold them together.
When would I use Deep scan?
When Words and Meaning have both come back empty and you are sure the moment is in a particular folder. Deep scan reads every line in scope rather than ranking an index, so it is slower and it is meant to be pointed at a narrow scope first.
What are connected moments?
When you open a passage, the related passages elsewhere in the project are listed beside it, each with a sentence on what connects them: the same event told from another chair, a phrase that recurs, a claim that another speaker contradicts. They are written by a model reading your indexed transcripts, and the feature can be switched off per project.
Related capabilities
Further reading
Background guides and comparisons.
Embedding-based semantic search finds passages by meaning rather than by matching words. Here is what an embedding is in plain terms, how the search works, and where it still misses things.
Retrieval-augmented generation answers questions about indexed interview transcripts by retrieving relevant chunks and grounding the model's output in those chunks. Here is how the pipeline works and where it still fails.