Inception is live on Pump.fun Go. Browse bounties

Embodied-AI training data

Teach robots a trade.

The next wave of robots will fold laundry, fix pipes, and work the line. So will the vision models and embodied agents behind them. They learn by watching the people who do the work, so we pay those people to film it.

Built on Pump.fun Go · Paid in USDC · 100% consent-verified

How it works

AI and robotics teams place bounties. We run the campaigns.

AI and robotics teams come to Inception for the POV footage they need. We collect the funds and manage the whole Pump.fun Go bounty campaign, sourcing it from the tradespeople who actually do the work.

01Demand

An AI or robotics team places a bounty

A company building robots or AI tells Inception exactly what POV footage it needs to train on, sets the spec, and funds the prize pool. One order, and they never touch the crowd.

02Managed

We run the Pump.fun Go campaign

Inception collects the funds and launches the bounty on Pump.fun Go on the company's behalf. We manage moderation, the crowd, and payouts end to end.

03Supply

Tradespeople submit POV

Workers who actually do the job film it first-person and submit on Pump.fun Go. Every entry is judged against the spec. The spec is the rulebook.

04Settle

Winners paid, footage delivered

Up to 5 winners get paid from the prize pool. The company gets the licensed POV footage, and Inception keeps a flat 10%.

What we collect

The jobs robots and AI can't do yet

Physical-labor robots and embodied AI are bottlenecked on real-world demonstration data from skilled, dexterous, on-site work. That's exactly what we source.

PLM

Plumbing

Leaks, fittings, water heaters, rough-ins, fixture swaps.

ELC

Electrical

Outlets, panels, wiring, lighting, troubleshooting.

HVC

HVAC

Installs, servicing, ductwork, diagnostics, refrigerant.

CLN

Cleaning

Residential and commercial turnovers, deep cleans.

DLV

Delivery & logistics

Last-mile routes, loading, doorstep handoffs, returns.

FAC

Factory & warehouse

Assembly lines, picking, packing, machine tending.

CAR

Carpentry & handywork

Framing, mounting, repairs, assembly, finishing.

APP

Appliance repair

Diagnostics, part swaps, installs, calibration.

We're planning to open up more trades over time. Want one you don't see here? Tell us what you need.

The format we buy

Shoot it from your point of view.

Strap a camera to your chest, hit record, and do the job. We want the work from your eyes and hands, the exact angle a robot or AI agent has when it does it. A camera on a tripod across the room is close to useless.

First-person GoPro view: a plumber's gloved hand pressing a copper pipe fitting up between floor joists, lit by a tool light.
REC 00:02:14
1080P · 30FPS
POV · CHEST-MOUNT
PLM-014
Real submission. Copper repipe, chest-mounted. This is the bar.
01
Wear the camera

Chest or head mount. Hands-free, roughly eye level (what the robot will see).

02
Keep hands in frame

Your hands and the part you're working on, in shot the whole time.

03
One take, start to finish

Roll before you start, stop when it's done. Don't cut the fiddly bits.

04
Real job, real mess

Actual work in a real space beats anything staged. 1080p+, hold it steady.

Don't bother sending
Vertical videoFilmed from across the roomCutting away mid-taskHands or tool out of frameCustomer faces / addresses

Gear's cheap. A phone in a $15 chest strap, a GoPro, or a clip-on bodycam. Film the job, then submit it on Pump.fun →

What you can do with it

Any volume of training data, for any model.

Tell us the task, and we source the raw first-person footage at the scale you need, for any AI agent or robotics team. Every clip is dense with what models learn from: the objects in frame, every step of the task, the hands and the motion. Label it with your own pipeline, or train straight on the raw video.

Imitation learningVision-language-action modelsWorld modelsManipulation policiesEmbodied agentsBenchmarking & evals
A first-person clip annotated the way a training pipeline sees it of a copper repipe, detecting the pipe, fitting, press tool, and the worker's gloved hand.
ANALYZING · FRAME 142
What models seeLIVE
Detected
copper pipe0.98
copper fitting0.94
CPVC riser0.90
press tool0.96
gloved hand0.99
Task

Press-fit a ¾″ copper joint

Step

Step 3 of 5 · seating the fitting

Hands2 trackedMotionpressDeliveredraw MP4 · 1080p
For AI and robotics teams

Place a bounty. We really get the videos.

Tell us the POV footage your models need and fund the prize pool. Inception launches and manages the bounty on Pump.fun Go, moderates every submission against your spec, and delivers the licensed footage. You place one order, we handle the crowd.

01
Place the bounty

Define the trade, the POV shot, and the prize pool. Your spec is the rulebook every submission is judged against.

02
We run the campaign

Inception collects the funds and manages the whole Pump.fun Go bounty: moderation, the crowd, and payouts. You don't touch any of it.

03
Up to 5 winners

The best POV submissions win and get paid from your prize pool. More entries, more coverage, better data.

04
You get the footage

Licensed, consented POV footage delivered to you. 90% of your spend reaches the workers, and our 10% is the only cut. Refundable if a campaign can't run.

Sourcing at scale or need a custom spec? Talk to our data team →

Start a bounty

Tell us the footage. Fund the pool.

$
Prize pool $45Inception fee $5 (10%)

Pump.fun Go pays up to 5 winners. More winners, more coverage.

Next, fund the prize pool in USDC. We launch and manage your campaign on Pump.fun Go and deliver the footage.

Done for you

Funds in, footage out

Send USDC and describe the trade in plain English. We connect the accounts, write the bounty spec, fund the escrow, and run the campaign on Pump.fun Go. You get a footage deliverable, not a crypto project to babysit.

Judged on merit

An independent reviewer picks the winners

A neutral third party scores every clip on what actually trains a model: true first-person POV, unedited, real work, no AI. We do not grade our own homework, and the best footage wins on quality, not on who posted first.

Guaranteed both ways

The loop closes every time

Buyers get usable, licensed footage, or a refund if a campaign cannot run. Up to 5 workers per campaign get paid from the prize pool, and the ones who deliver build a track record that wins them the next one.

For tradespeople

Your day on the job is the bounty.

Tradespeople, drivers, and factory workers: the skills in your hands are scarce and valuable. Film the job in POV, submit to a live campaign on Pump.fun Go, and win the prize.

  • Win a bounty, get paid in USDC. Each campaign pays up to 5 winners.
  • Film the work you already do: plumbing, wiring, HVAC, deliveries, the line.
  • Any phone or body-cam works. Chest-mount it, shoot POV, submit.
  • You stay in control of consent, your face, and the jobsite at all times.
Submit your footage

Live bounties run on Pump.fun Go. Film the job from your point of view, submit, and win the pool.

Browse bounties on Pump.fun Go

Get new-campaign alerts

We'll email you when a bounty opens in your trade.

$ICPT is not live yet. Launch will be announced here first.

The $ICPT token

Every contract buys it back,
and holds it.

Real companies pay for real footage, and 5% of every contract buys $ICPT into the community treasury. Owned by the network.

Every contract
5% buyback to treasury
Fair launch
Pump.fun Go on Solana
Backed by
licensed footage

$ICPT is a community and utility token. Not a security, not a promise of profit. Crypto is volatile, do your own research.

INCEPTION · NETWORK · SOLANA$ICPT
FAQ

Questions, answered

Inception sources point-of-view video of real work and turns it into training data for robots and AI. An AI or robotics team tells us what footage it needs, we run a bounty for it on Pump.fun Go, the people who actually do that work film it from their point of view, and we deliver the licensed footage, ready for your training pipeline.

Robots, AI agents, and embodied systems are about to take over the boring, manual, repetitive work: folding laundry, fixing pipes, running the line. They learn that work mostly by watching people do it, and real first-person footage of hands-on jobs is scarce and expensive. That footage is the bottleneck, and it is what we source.

Place a bounty: pick the trade, describe the shot you need, set a prize pool and how many winners (up to 5), and fund it in USDC. You never touch Pump.fun. We launch and run the whole campaign, moderate every submission against your spec, and deliver the licensed footage.

Yes. The same first-person footage trains vision-language-action models, world models, multimodal vision models, embodied agents, and evals, not just robot policies. If your system needs to understand how people do real, hands-on work, place a bounty the way a robotics team would and we run it end to end.

Raw, clean, consented first-person clips of the work you asked for, in high resolution. They are dense with what models learn from, the objects, the steps, the hands, the motion, so they feed straight into imitation learning, vision-language-action models, manipulation policies, or evals, with your own labeling tools or none at all. Need a specific format? Tell us.

Yes, any volume, for any AI agent or robotics team. Start with a small test bounty to check the quality, then scale the prize pool to pull more submissions across many trades, hands, and real sites. The whole model is built to source data at whatever scale you need.

You fund a prize pool. 90% goes to the people who do the work. Inception keeps 10%, split evenly: 5% buys back the $ICPT token into the community treasury, and 5% funds the team. No subscription, no seats, you pay per bounty.

It is held until the campaign runs. 90% pays out to the winning submissions (up to 5) in USDC, and 10% is our fee. You see the split before you pay, and if a campaign cannot run, your funds are refundable.

No. You place one order and pay one price. We handle the on-chain mechanics, the Pump.fun Go campaign, the moderation, and the payouts. You just receive the footage.

It is open to anyone, no application. Film a real job from your point of view and submit it to a live bounty on Pump.fun Go. The best submissions win, and each bounty pays up to 5 winners from the prize pool, in USDC. Better POV, better odds.

In USDC on Solana, through Pump.fun Go, once submissions are judged against the spec. It is a stablecoin, so winnings do not swing with the market, and there are no minimums.

$ICPT is backed by the footage contributors create and the companies that pay to license it. 5% of every contract buys it back on the open market and holds it in a community treasury, so real demand flows into the token instead of hype. It is a community and utility token, not a promise of profit. There is a full breakdown on the token page.

Only film what you are allowed to. Many jobs need just your hands and the work, not faces or addresses. We support face, document, and background blurring, and you confirm you have the right to film the site. Never record a customer's private space or a proprietary facility without consent.

Only submit footage you have the right to share. If your employer or client owns the worksite or the work product, get their permission first. You keep ownership of what you submit and license only the clips you choose. You grant no rights to anything you do not submit.

The team that funds the campaign licenses the winning clips, with consent recorded and on-chain provenance for every one. Terms are set when the bounty is placed, so there are no surprises for buyers or submitters.

Yes. AI and robotics teams can place a bounty today, and tradespeople can submit to live bounties on Pump.fun Go. We are opening trades in waves, and anyone, anywhere, can take part.

Still curious? hello@inception.training

AI and robots are coming for the trades

Capture the work.
Own the dataset.

Commission the footage you need, or get paid to record the job you already do. Drop your email and pick a side.

One email when it matters. Unsubscribe in one click.