NEWMVS for embeddings. Managed for files. Both on object storage.Vectors or files. Pick a path.Start →

Where we lead

Find what makes your best ads work

The hooks and patterns in your top-performing video ads, ready for the next brief.

Brand marks, faces and policy risks flagged in every segment of every video.

Get answers from documents

The answer out of every PDF and scan, with the page it came from.

Most visited

Search game film

Describe the play and get timestamped clips back from every game.

Organize a media library

Search and license every film, episode and clip by what is seen and said.

See them running in the showcase →

All solutions →

Look it up against data we host

Screen a frame for IP risk

Every public figure and brand mark in one frame, scored.

Find trademarks in a frame

The registered brand marks it looks most like, ranked for review.

Check copy against trademarked slogans

The registered slogans a line resembles, and who owns them.

Identify public figures

Every public figure a photo matches, with a confidence score.

Classify content for ad targeting

IAB content categories for a page or transcript, across all four tiers.

All retrievers →

Deploy on your own files

Find your best ad creative

Every scene of every ad, and the ones that keep earning spend.

Segments, keyframes and brand-mark matches for a moderation policy.

Search footage archives

Millions of scenes, clustered so you can move through them.

Match video to articles

An article in, a ranked playlist of video out.

Search UX sessions

Transcripts joined to what was on screen, across every session.

Monitor live cameras

RTSP feeds you can search, with alerts from retriever results.

Curate training data

Turn model failures into training and evaluation sets.

All templates →

Run open models on your data

Text, image and video vectors for search.

Objects located in images and video frames.

Pixel masks for the things in each frame.

Video and image Q&A

Ask a question about a frame or a clip.

Sign in Sign up

Reduce

Summarize

Condense multiple documents into a summary using an LLM

summarize

Note: This playground provides simulated output to showcase functionality. No data is processed on our servers. Use this demo to explore the stage's configuration options before integrating it into your retriever pipeline.

Configuration

modelstring

LLM model to use for summarization.

max_lengthinteger

Target maximum length of summary in words.

stylestring

Summary style: 'concise', 'detailed', 'bullet_points'.

promptstring | null

OPTIONAL. Custom prompt template for summarization.

Output

No output yet

Configure the stage parameters and click "Run Stage" to see the simulated output.

Related to summarize

Pages on the same stages, extractors and patterns.

Aggregate

Compute aggregations (COUNT, SUM, AVG, etc.) on pipeline results

Group By

Group documents by field value for decompose/recompose workflows

Cluster

Cluster documents by embedding similarity to discover themes

Sample

Sample a subset of documents using random or stratified sampling

Temporal

Group documents by time windows and compute per-window aggregations with drift detection

Limit

Truncate results to a maximum count with optional offset for pagination