Human Archive

First-person video, depth, motion and touch data of real work for robot labs, from head-worn rigs and a phone app.

No public API listed

Mode
Made to order · Speed not stated
Coverage
India, United States +1
Price
On request
Key
Required
Resale
Not stated
AI training
Not stated

Eight links

1 of 8 links claimed · 0 evidenced

Compiled from public sources, unclaimed. What they mean

Each link
  1. 1. Real personClaimed

    Person · Claimed. App contributors hold accounts, confirm they are 18 or over, and pass identity checks run by the payout providers (Stripe, Circle) where required; Human Archive receives only the verification status. Workplace footage comes from workers at partner businesses.

    Still open: No identity method is published for partner-business workers, and no per-recording identity record is public.

  2. 2. Captured locallyNot stated

    Camera · Not stated

    Still open: Not a signed capture, so none (STANDARD §3). What is stated: Global-shutter stereo cameras with a hardware-synchronised IMU on the rig; phone camera, IMU and LiDAR in the app. Contributors confirm each recording is genuine and not altered with generative-AI or post-production tools. No signed capture; Human Archive is not on the C2PA conforming products list. Screen or generated footage is excluded only by contributor confirmation and review.

    Place · Claimed. Collection metadata includes the environment and location; the app uses device location only if the contributor grants permission.

    Still open: No location proof is documented, and app location is optional.

    Needs both: the weaker one counts.

  3. 3. Genuine deviceNot stated

    App · Not stated

    Still open: Workplace capture uses Human Archive's own head-worn rig and app tasks are recorded in its app (nothing uploads until Submit), but no device or app attestation is documented. Capture in the provider's own app or rig is not a server-verified attestation (STANDARD §3).

  4. 4. Sealed captureNot stated

    Camera · Not stated

    Still open: Not a signed capture, so none (STANDARD §3). What is stated: Global-shutter stereo cameras with a hardware-synchronised IMU on the rig; phone camera, IMU and LiDAR in the app. Contributors confirm each recording is genuine and not altered with generative-AI or post-production tools. No signed capture; Human Archive is not on the C2PA conforming products list. Screen or generated footage is excluded only by contributor confirmation and review.

    Content · Claimed. App clips: on-device hand-visibility check, about one frame a minute sent to Google's Gemini API to confirm the assigned task, human review of flagged clips and random audits. Workplace footage: a QA model cuts and filters sessions, then annotators verify clips and labels and an independent reviewer approves them (100+ annotators; 0.763 inter-annotator agreement on boundaries).

    Still open: Acceptance rates and per-dataset QA results are not published beyond Ego500's agreement figure. Frames sent to Gemini may be reused by Google to improve its own products (privacy policy §4.3).

    Counts only with a stated signing mechanism (C2PA, a manifest).

  5. 5. Known timestampReported

    Time · Claimed. Session metadata records when a recording was collected; rig video and IMU share a hardware clock.

    Still open: The hardware clock aligns streams to each other; no signed or independently time-stamped capture time is documented.

    A time stated, not attested (RFC 3161, a log): Reported.

  6. 6. Tracked in transitNot stated

    Custody record (optional field) · Not stated

  7. 7. Independently verifiedNot stated

    Cross-check · Not stated

    Still open: Nothing independent confirms individual captures. HARD-1 and HARD-2 are Human Archive's own robot benchmarks of data usefulness, not checks of provenance.

  8. 8. Delivered to youNot stated

    Delivery receipt (optional field) · Not stated

API

Needs a sales or approval step first.

No public API listed.

Sources

18 sources
  1. Data and deployments infrastructure for physical AI; works with AI labs to scale human video/sensor data; OTS catalog; contact form and cal.com booking; [email protected] humanarchive.ai
  2. Buyer contact form fields (name, email, notes); 'researcher dashboard and samples' humanarchive.ai
  3. Contributor licence (perpetual, irrevocable, non-exclusive, sublicensable; for ML and robotics training; included in datasets licensed to customers); contributors keep ownership; payout only on approval; 18+; not offered in IL, TX, WA, OR; approved clips de-identified and kept on deletion; video tasks have no audio humanarchive.ai
  4. Per-session consent confirmation logged; metadata on how, when and where recorded; depth, IMU and calibration data; Gemini frame checks; Stripe and Circle payout and KYC status; customers licensed under written agreements restricting use, banning re-identification, requiring confidentiality; optional location permission humanarchive.ai
  5. Recording rules: only own hands and task, no other people or children; employer written permission at work; genuine, unaltered recordings; on-device, AI-frame, human and random-audit review humanarchive.ai
  6. Voice Exchange: audio-only, one-time and per-call consent, separate voice datasets, distributor-network contributors paid by the network humanarchive.ai
  7. App recording flow (double-press volume button; nothing leaves the device until Submit); payouts after approval humanarchive.ai
  8. iOS app by Human Archive, Inc.: pays for approved contributions; payout shown on each task; version 1.2.0 apps.apple.com
  9. Ego500: 500 hours across 60+ work environments; custom multi-camera hardware deployed in workplaces; metadata including location; QA model, annotator verification, EgoBlur face blurring, independent reviewer; 100+ annotators; 0.763 inter-annotator agreement ego500.humanarchive.ai
  10. Head-worn rig: global-shutter stereo pair with hardware-synchronised IMU and shared hardware clock ha-slam.humanarchive.ai
  11. HARD-2: 100 hours from the OTS catalog; self-run robot benchmark; collection rigs multi-camera, 60 fps, global shutter humanarchive.ai
  12. HARD-1: egocentric stereo video with 3D hand pose and world-frame SLAM; self-run benchmark humanarchive.ai
  13. $8.2M seed; 1,000+ active headsets; 50+ devices; tactile gloves, mocap suit, wrist cameras with RGB-D; $1/hour base pay; customers choose a discounted recorded visit or a full-price unrecorded one; unclear what workers are told; India first, expanding to Southeast Asia and US pilots techcrunch.com
  14. 120+ partnerships, mostly India; partners contractually responsible for informed consent; '$1-10 per hour' average (unverified by Inc42); tens of thousands of hours collected; Snabbit controlled-environment test and NDA inc42.com
  15. Wing VC led the $8.2M seed; rigs capture RGB, audio, stereo depth, tactile, mocap, wrist cameras; network in the U.S., India and beyond; data ships to robotics labs www.wing.vc
  16. No Human Archive server in the official MCP Registry registry.modelcontextprotocol.io
  17. Human Archive not on the C2PA conforming products list raw.githubusercontent.com
  18. Privacy §4.3: frames sent to Gemini API covered by terms permitting the provider to use submitted data to improve its products; §6.1 customer agreements restrict use to agreed purposes humanarchive.ai
Notes

About. Human Archive (Human Archive, Inc., San Francisco; $8.2M seed, May 2026) collects first-person data of real work for robot learning. Workers at partner businesses, mostly in India, wear Human Archive's head-mounted rigs (stereo cameras, IMU, plus tactile gloves, mocap suits and wrist cameras), and paid contributors record tasks in its iOS app. It processes the footage into 3D hand pose, SLAM poses and dense action labels, and licenses datasets, from an off-the-shelf catalogue or collected with labs, to AI and robotics companies under written agreements. Prices, turnaround and the buyer licence are not published. Consent is published for app contributors but not for workers recorded through partner businesses.

Price. No prices are published. Buyers use a contact form ('we'll send over our researcher dashboard and samples'), book a call, or email [email protected].

Coverage. Press reports put most collection in India, through 120+ partner businesses (hotels, restaurants, home services, quick commerce), with expansion to Southeast Asia and early US pilots (TechCrunch, Inc42, May 2026). Wing VC describes a network across homes, factories and warehouses in 'the U.S., India, and beyond'. The contributor app is on the US App Store; its terms exclude residents of Illinois, Texas, Washington and Oregon. Human Archive's own site lists no countries.

Timing. Human Archive says it is 'working alongside leading AI labs to scale human video/sensor data' and offers an 'OTS catalog'; it publishes no turnaround for new collection.

Licence. App contributors keep ownership of their clips and grant Human Archive a worldwide, perpetual, irrevocable, non-exclusive, sublicensable licence to use them 'to develop, train, evaluate, and improve machine-learning and robotics models' and to include them 'in datasets Human Archive licenses to its customers for those purposes'. Customers receive data 'under a written agreement' that restricts use to the agreed purposes, prohibits re-identifying contributors and requires confidentiality (privacy policy). The buyer licence is not public; resale, buyer ownership and Human Archive's reuse of a buyer's data are not stated. Terms for footage collected through partner businesses' workers are not published. The Ego500 release is described as open-source, but no licence is shown on its page. No buyer licence is public, so whether a customer may train AI is not stated by the provider; training is the evident purpose but is set by each contract.

Compiled from public sources. Is this your company? Publish your card ›

Compiled from public sources · updated 2026-10-28JSONSiteCompare