Humanbased helps AI teams launch accountable human, expert, and agent-powered data workflows with attribution, compensation, and reusable data products built in.
Commission AI training data, evaluations, real-world feedback, and frontier data campaigns. Qualified contributors and teams get paid upfront, keep attribution, and can share in future value when their work is reused.
Campaigns are running now in invite-only beta. General availability opens once payout and lineage flows hold up in production.
We are converting the 10M+ Codatta signup base into verified contributors, real campaign users, and accountable paid work.
Buyer briefs and contributor qualification.
First campaigns with close operator review.
Invite-only access expands to more campaigns and contributors.
Public access after payout and lineage flows are stable.
Loading yesterday's payload...
Contributors qualify for paid work with attribution. AI teams commission campaigns.
For qualified people, experts, self-forming teams, and managed groups that want paid AI data work with clear review.
See how it works for contributors. Access is invite-only during beta and opens further at general availability.
Contributor portalFor AI labs, startups, and agent builders that need training data, evaluation, expert review, or frontier campaign output.
See how it works for developers. Docs are open; for a guided pilot, write to hello@humanbased.ai.
Developer portalHow it works for developers How it works for contributors The full operating model
Developer-authored workflows for sourcing, labeling, validation, evaluation, and dataset assembly. Mix humans, experts, teams, agents, and compute services in one campaign.
Serve AI labs, startups, agent builders, independent experts, self-forming teams, and company-led teams with clear roles, quality signals, payouts, and reputation.
Each accepted contribution can carry provenance, usage rights, attribution, valuation history, and a path to future revenue when data access or ownership changes hands.
Physical interaction traces, embodied task demonstrations, environment annotations, failure cases, and simulation feedback for physical AI and video generation teams.
Licensed, consented speech with speaker provenance, accent coverage, emotion labels, dialogue review, and pronunciation QA.
2D-to-3D assets, design review templates, concept-match scoring, and expert QA with licensing clarity attached.
Benchmarks, rubric design, and human judgment from calibrated expert cohorts, reusable across evaluation rounds.
Accepted work does not disappear into a dataset. It stays attributed to the people and teams who made it, with usage rights and valuation history attached.
Contributors are paid upfront, primarily through 
USDC, and lineage
can anchor onchain through 
Base and 
Ethereum via XnY so
attribution, access, and royalty eligibility travel with the data
after it leaves Humanbased.
The monorepo is public and the research is published. Recent writing from the team:
What forty real-world data deals and nine pre-registered experiments say about when royalties are the right way to pay for data. Research, July 2026.
A retrospective of two months of monorepo PRs, with findings on PR shape, review cost, and model routing. Engineering, June 2026.
Most AI data buying splits into three motions. The difference is the operating model, not the vendor logo.
Brand Kit
Logo files, usage rules, color tokens, typography, and export assets for partners and builders.
