Food and Beverage social data for agents

Give agents eyes on social media. We watch every food video on social and turn it into data any agent can read.

Available forInstagramTikTok
1 Watch the reel
0:08
2 Write down what happens
3 Turn it into rows 0 of 14
atwhattagvalue

What’s in the data

ingredients13.6Kevery food named or shown
dishes27.2Kwhat they end up in
brands and products47Kand counting
kinds of tag40taste, occasion, pairing, verdict…

A tag is one thing a post says about a food: what it is made with, how it is cooked, what people think. Every tag carries the quote and the second of video it came from.

Use cases

Three jobs the same tables do, with an example report for each.

Brand tracking and social listening

Every time your brand or a competitor's shows up in a food video, with what was said about it and whether it was positive. From the video, not the caption.

See the example report

Ingredient and F&B insight

What an ingredient is being made into, what it is paired with, how it is prepared, and the verdicts it gets.

See the example report

Food trend detection

Which dishes and ingredients are rising week over week, with the posts that prove it and the creators behind them.

See the example report

How it’s made

Nobody writes the categories in advance. We watch the video, the taxonomy grows out of the posts, and the trend layer counts only what clears the bar.

01

Source

Food creators on Instagram and TikTok, found by what they post, not by hashtag. A creator stays in the panel only while most of their posts are food.

02

Watch

Every post is read four ways: the caption, the transcript, the on-screen text, and a timestamped account of what happens in the video.

03

Tag

Each thing a post shows or says becomes a tag: the food, what was said about it, positive or negative, with the quote. New foods are proposed, checked against the taxonomy, then added or merged.

04

Deliver

Daily files into your bucket, or live to your agent over MCP. Every row with its quote and the second of video it came from. Per-item releases add share of conversation over time, with confidence intervals and a floor.

05

Deploy

For brands, we do this with you. We load the feed into your database or Claude Code over MCP, hold coverage against what you use today, and build the first report alongside your team. Written collection criteria before day one.

Book a demo

What you receive

Everything we make from a video lands in nine tables. Pick one to see its rows for the video at the top of the page.

Tables9
posts · rows for the video at the top10 rows shown
columnthis reel
platforminstagram
posted_at2026-09-01 11:03 ET
like_count2,184,797
comments_count12,009
media_typeVIDEO
formatEducational · Tutorial
regionNORTH_AMERICA_NON_MEXICO
subjectZero-waste cooking using leftover rice
captionZero waste rice 🍚 #rice #zerowaste #leftovers
play_by_play0:23 · She says ‘And now my favorite’. She blends rice with peas, chives, and water, rolls the paste thin, dries it, and fries it.
all columnspost_idplatformpermalinkcaptionmedia_typelanguageposted_atweekview_countlike_countcomments_countshare_countsave_countis_adcreator_idtaggedsubjectsummarykeywordstopicsformat_categoryformat_subtyperegionbrands_identified
SELECT * FROM 'evidence/posts.parquet' WHERE permalink LIKE '%Dcv8EBzRVFL%';

One row per post. The caption, the transcript, the on-screen text and the play-by-play ship as passages in post_text.

Replaces a raw social feed, a tagging pipeline, and the analyst hours on top.

Where it lands

Your stack

Parquet into your S3 or GCS bucket, or rows into the database you already run. Every morning: new posts, updated tags, new metric observations.

Your agent

An MCP server over the same tables (MCP is how agents like Claude connect to tools), so your agent asks the data directly.

works withparquet · rows · mcp
ClaudeMCPAIOpenAIawsAWSGoogle CloudAzAzurePostgresBigQuerySnowflakeDatabricksDuckDBsociable data

Common questions

Where does the data come from?

Instagram and TikTok, collected directly from the platforms. The collection method ships as a written document with every release.

How do I know your panel covers my category?

Ask for a one-day slice before you buy. We'll run coverage against your own creator list, by follower tier.

How do you deliver it?

Daily Parquet files to your S3 or GCS bucket, rows into your Postgres, or an MCP server your agents query directly. Three months of history standard, twelve on request.

What's in a row about a creator?

Followers, age band, gender, region, each with its evidence. No race, ethnicity, accent labels or embeddings.

How is it priced?

Per post, per platform, on a committed volume. Everything is included at every tier.

Can I resell it inside my product?

Yes. Scoped in the licensing agreement.

See your own items in the data

Tell us three ingredients, dishes or brands you track. We pull one day of posts on them and walk you through the rows on a 30-minute call.