The open video index
All the world's video, searchable.
Your agents need to see and hear the world.
We're building the open index that lets them.
Your agent watches hours of video so you don't have to. You get the one moment that answers your question, with a link to the creator who made it.
Give your agent eyes
Context is everything.
Video has the most of it.
Video bridges the real world and the digital one. If your agents can't watch what people post every day, they're missing what's actually going on.
What happens in real life gets filmed and posted, on every platform, every minute. It's how people react, share their moments and decide what to buy, where to stay and who to trust. A page of text is a summary; the video is the thing itself, with far richer context: what was shown, how it was said, who was there. Agents won't just read pages. They'll need to see and hear, and that matters more every year.
We believe in video that much, so we're building the open video index for every agent, across every platform. Help us build it.
- Big platforms
- Index their own video. Keep it.
- Everyone else
- Every agent pays to process the same clip again.
- UpRock
- Indexed once, searchable by every agent. Shared and open.
Two ways this goes
Who should own the video index?
Video is where the real world shows up online. We're already filming everything: stories, reactions, reviews, the street outside, the meal in front of us. An agent that can't watch is blind to all of it. Someone will build the index of it. The question is who owns it.
Closed
- A few big companies pre-index everything and keep it.
- You pay a lot, and your data goes to them.
- One platform at a time.
Open
- One public index of public video, kept by a community network.
- Low prices, and the data is published for anyone to use.
- Across platforms, and every video an agent pays to index joins it.
For agents, today
Find the exact second in a video.
Building an agent? Every agent needs to know what's happening right now, and that's on video. One MCP server, one key, and your agent can watch.
How video semantic search works, with a live demo →# MCP tool on https://mcp.uprock.ai/mcp (Authorization: Bearer $UPROCK_API_KEY)video_search({ "query": "rooftop pool hotel lisbon" })# → moments: video URL, start/end time, transcriptvideo_moment({ "url": "<video url>", "t0": 312, "t1": 342 })# → what is said and shown in that rangePricing
- Search$1/ 1K queries
- Index a new video Private beta$0.02/ minute
No storage fees. UPT is always 20% less. What you index joins the shared index for everyone.
Indexing new video is in private beta and opens up soon. Contact us for early access →
1
Search it
Ask for a moment in plain language. Get the video, timestamp and transcript.
2
Feed it
Pay to index a channel, playlist or link. It joins the shared index for everyone. Private beta: contact us for early access.
3
Earn with it
Run a Rocket to earn UPT, then swap it for credits and pay 20% less.
Built by the network
Indexed once.
Searchable by every agent.
UpRock is a DePIN, a decentralized physical infrastructure network: the devices belong to our community, not a data center. It's how we can offer real-device crawling at the price we do, and now we're doing it for video. Idle Macs owned by our community do the slow work of watching and labeling. The moment an agent asks, our servers answer. How the Fuel Engine works.
1 · Network
Fetch
Rockets load public video from real devices, the way a viewer would.
2 · Macs
Watch
Sample frames every few seconds, and gather the captions and metadata the video already has.
3 · Macs
Label
Describe what is said and shown in each moment: who, what, where, any text on screen.
4 · Macs
Embed
Turn every 30-second moment into something an agent can search by meaning.
5 · Our servers
Search
Answer an agent in real time with the video, the timestamp and the transcript.
Why this matters
An agent that can't watch
is blind to most of the web.
01
Creators get found
There is more video than anyone can watch. An agent that can watch finds the exact moment a person asked for, from the creator who said it best, and sends them there.
This isn't scraping for copies. We keep notes on each moment and link back to the source, so the views go to the creator.
02
Agents see what is real
More and more video is made by AI. A transcript and a couple of frames can make a fake and a real clip look the same.
Agents will need to see what is shown, not just what is said, and check it against other tools: community notes, the news, other sources. Video is the richest context an agent can get about the world and about what you mean when you share one.
03
Decisions come down to a moment
People think in stories. We picture the trip, the meal, the car before we choose it, and the final evidence is often a clip: the review, the test drive, someone we trust saying it.
If your agent can't watch, it can't send you the moment that helps you decide.
In development · Early access
Got a Mac mini?
Bring your Mac mini and help build the world's largest open video index. Your Mac does the indexing while you're away and earns UPT for each hour of work it completes. Early machines get in first.
- GathersCollects each video’s metadata, captions and source: title, creator, platform, link.
- WatchesDescribes what’s on screen in each moment: people, places, products, text.
- IndexesTurns every moment into something any agent can search by meaning.
It's patient work: a video can be indexed over hours, not milliseconds, so it runs while your Mac is idle. When an agent searches, our servers answer instantly. Decentralized crawling, compute and inference, on machines people own.
Any Apple Silicon Mac with 16 GB or more counts. Got a Windows or Linux machine with an NVIDIA RTX 30-series card or newer? Tell us that too.