Briefed against your spec
The target is agreed in writing before a single contributor is briefed, so assignments describe exactly the tasks, environments, and edge cases you asked for.
Our contributor network

How the network works
The target is agreed in writing before a single contributor is briefed, so assignments describe exactly the tasks, environments, and edge cases you asked for.
Contributors are paid hourly, including setup and retakes. Payment is not contingent on acceptance, which removes the incentive to pad volume with weak recordings.
Assignment, upload, and review run through one web portal, so every recording is tied to an assignment, a contributor, and a review decision.
Recordings that miss the spec, or where a bystander cannot be de-identified, are dropped and logged, not shipped at a discount.
Overview
A provider that runs its own contributor network can answer the questions a buyer’s counsel and a model team both ask: who recorded this, what did they agree to, what were they paid, and why was this clip accepted? A provider that resells or scrapes data generally cannot.
It also means the data is collected for your model. Nothing is gathered speculatively; contributors are briefed only once the spec is agreed.
Workflow
Your written spec is turned into assignments in our contributor portal. Contributors are briefed, sign a release, and capture against the assignment, using a specialized rig such as our synchronized egocentric hardware where the project needs it. Recordings are uploaded through the same portal.
Uploads missing camera telemetry are rejected automatically. Every remaining recording is human-reviewed for task match, visible hands, and staged or idle footage before it is accepted.
Contributors
Contributors are people who apply to capture for Firsthand in their own town and are briefed on each assignment. The release they sign is published verbatim on the provenance page and the for-collectors page links to it, so they know what they are agreeing to before they apply.
They are paid the hourly rate stated in their schedule for all time in session, including setup, calibration, and breaks, with no per-task bounty and no rejection clawback. They may withdraw consent for future sessions at any time and request deletion of any recording not yet delivered to a licensee.
Privacy
Contributors confirm they have not knowingly recorded anyone who has not signed a release. At ingest, faces, licence plates, and screens are detected and irreversibly blurred, then confirmed by a reviewer.
If a person cannot be de-identified, the clip is discarded rather than shipped.
Delivery
Accepted recordings ship with metadata, consent records, and the reject log, in a human-readable, PII-free file layout. Custom collections deliver in RLDS, LeRobot, and WebDataset, plus HDF5, zarr, and Rerun .rrd.
The data is buyer-owned under a perpetual, irrevocable, worldwide licence that is non-exclusive by default, with a data card recording the jurisdictions of capture and the consent basis.
Timing
Where coverage for your tasks and regions is deep, the median from signed spec to first validated batch is a median of 48 hours. Where it is thin, turnaround runs three to four weeks while recruiting expands in parallel across more regions.
A Spec Pilot, a small, fast first batch, validates the spec and accept criteria before you commit to volume.
At a glance
Every collection runs through the same seven-stage end-to-end custom collection pipeline. End-to-end custom data collection is a managed service that takes an AI data need from problem to owned dataset in a single accountable pipeline.
Network workflow
| Stage | Who does it | Control in place |
|---|---|---|
| Assignment | Firsthand, from your written spec | Nothing briefed before the spec is agreed |
| Consent | Contributor, before capture | Signed release, published verbatim |
| Capture | Briefed contributor | Hourly pay, not contingent on acceptance |
| Upload | Contributor, via the portal | Missing camera telemetry rejected automatically |
| Review | Firsthand reviewers | Human review of every recording; reject log |
| De-identification | Firsthand, at ingest | Faces, plates, screens blurred; unresolvable clips discarded |
FAQ
Tell us the tasks, environments, and streams your model needs. We will quote reach, timeline, and price, and brief our own contributors once the spec is agreed.