NOW LIVE · WEB, API, MCP

Video in.
Your JSON out.

Upload a video. Hand us a JSON schema. Get back exactly the structured data your workflow expects. Nothing more, nothing to configure.

30 free minutes. No card required.

standup-2026-08-04.mp412:38

04:12

RESPONSE200 OK

{
  "meeting_title": "Q3 pricing review",
  "participants": ["Ada", "Ravi", "Mei"],
  "decisions": [
    { "at": "04:12", "text": "Ship usage tiers" }
  ],
  "action_items": [
    { "owner": "Ravi", "due": "2026-08-11" }
  ],
  "sentiment": "aligned"
}

Eadweard Muybridge, Animal Locomotion Plate 626 (1887): a horse and rider photographed across a grid of sequential frames.

FRAME 07

1887 / ANIMAL LOCOMOTION, PLATE 626

Cutting motion into frames is not the hard part any more.

Muybridge needed a row of tripwired cameras to turn a gallop into sixteen numbered frames. Frames are cheap now. Turning them into something your code can read is the part still worth paying for.

01 / THE ONLY THING YOU CONFIGURE

You write the schema.
We fill it in.

No prompt engineering, no extraction presets, no pipeline to assemble. Give us the shape you want and every response conforms to it: validated, typed, and ready to hand straight to your code.

YOUR SCHEMA

{
  "clip_summary":  "string",
  "speakers":      ["string"],
  "key_moments": [
    { "t": "string", "label": "string" }
  ],
  "topics":        ["string"],
  "sentiment":     "positive | neutral | negative",
  "brand_safe":    "boolean",
  "has_speech":    "boolean",
  "duration_sec":  "number",
  "language":      "string"
}

Plain JSON Schema, or just sketch the shape you want.

Conformance, not best effort

Every response is validated against your schema before it leaves us. Wrong shape means retry on our side, not yours.

Every field can carry a timestamp

Put a timestamp in your schema and it comes back with the data, so you can jump straight to that point in the video.

Change the schema, not the plumbing

Need a new field next week? Add it to the schema and re-run. There is no model to retrain and no config to migrate.

02 / INTEGRATE

One POST. No SDK to learn.

Send a file or a URL alongside your schema. Take the result synchronously for short clips, or hand us a webhook for long ones.

curl https://api.fovea.run/v1/extract \
  -H "Authorization: Bearer $FOVEA_KEY" \
  -F video=@standup-2026-08-04.mp4 \
  -F schema='{ "decisions": [{ "at": "string", "text": "string" }] }'

SHORT CLIPS COME BACK LIVE

Short clips return on the same request. No job IDs to poll for the common case.

WEBHOOKS FOR THE REST

Pass a callback URL for anything longer and we post the JSON to you when it lands.

MCP FOR AGENTS

Point any MCP client at our server and your agent gets an extract tool it can call on its own.

03 / THREE WAYS IN

Same engine. Whichever door you walk through.

One account, one key, one billing meter across all three.

WEBno code

Drag a file in, get JSON back

Paste your schema into the editor, drop a video, copy the result. Useful for one-offs and for proving the shape before you wire it up.

APIone POST

One endpoint, no SDK required

Send a URL or upload, send a schema, poll or take a webhook. If you can write curl, you are already integrated.

MCPfor agents

Your agent gets eyes

Point Claude, Cursor, or your own agent at our MCP server. It decides the schema it needs and reads video on its own.

04 / PRICING

One number: minutes of video.

No seat licences, no per-field charges, no separate line for the API and the MCP server.

The meter is video length. That’s it.

Schema size and field count cost nothing extra. However complex the shape, the price is the same.

12 MIN 38 SEC OF VIDEO

MEASURED PER SECOND= $0.63

Free

START HERE

Every surface, no card

30 min

Web, API and MCP included
Videos up to 30 minutes
Minutes do not expire

Minutes

MOST TEAMS

Prepaid minutes, top up when you run out

$0.05/ video minute

Top up $5, $25 or $100. A dollar is 20 minutes.
Measured per second, one minute minimum
Webhooks and batch endpoints

Volume

For sustained high volume

Talkto us

Committed-use discount
Region pinning and retention terms
Shared Slack channel

05 / BEFORE YOU ASK

The questions we get first.

Anything else, mail us. A person answers.

Which model is behind this?: Google's Gemini multimodal models today. The layer on top is ours and it is model independent, so when a better video model ships you get it without touching your schema.
What happens to my video?: Deleted once the job finishes. Nothing you send is used to train a model.
How long can a video be?: Thirty minutes per video at launch. Longer files come back on a webhook rather than on the original request. We raise the cap when someone asks for it.
What if it gets something wrong?: Shape is guaranteed. Judgement is not. Ask for a timestamp on the fields that matter and you can spot-check them against the video.
What happens when I run out of minutes?: Jobs stop rather than quietly billing you. Top up $5, $25 or $100 whenever you want, at twenty minutes to the dollar. Unused minutes carry over, nothing renews, and there is no subscription.

THIRTY FREE MINUTES · NO CARD

Make your agents video-aware.

Create a key and run your first extraction in the next five minutes.

Video in.Your JSON out.