Skip to main content
We’re building market infrastructure for AI training data. Reppo lets anyone publish content to domain-specific data markets called datanets, where stake-backed voters signal quality and contributors earn REPPO emissions. Rather than purchasing raw labor hours, Reppo creates open economic competition — aligning the interests of data publishers, curators, datanet owners, and the AI teams that consume the output.

​
What is an RL environment?

An RL environment defines the task, scoring function, and feedback loop used to improve a model. It is the setting where a model acts, gets evaluated, and learns. On Reppo, datanets act as market-based environments for generating and curating that learning signal.
The RL training loop: a task prompt feeds a model, a verifier scores the attempt, and the score and gradient feed back into the model across millions of rollouts
“Scale AI coordinates training data pipelines with contracts. Reppo coordinates them with markets, access controls, and on-chain incentives.”
By putting stake behind both contribution and curation, Reppo is designed to address the AI training data trilemma.
The AI training data trilemma: quality, scale, and legal
Reppo data markets: publishers to markets to curators to models, with economic stake on both sides, prediction-market style curation, and on-chain provenance

​
Key concepts

Datanets are domain-specific prediction markets where data contributors publish work and put capital behind its quality. Each datanet defines its own rules: who can publish, what content qualifies, how fees are set, and how rewards flow. Pods are the atomic data units inside a datanet — a tweet, image, video, annotation, or any media the datanet accepts. Each pod is minted as an on-chain NFT on Base and earns REPPO emissions epoch by epoch based on how well it performs in curation voting. veREPPO is voting-escrow REPPO. Voters lock REPPO tokens for a chosen duration to receive veREPPO, which gives them curation authority. Larger locks and longer durations yield more voting power per token. Splitting tokens across multiple identities does not multiply power, making Sybil attacks economically ineffective. Epochs are 48-hour windows during which publishers submit pods, voters allocate veREPPO, and rewards accrue. Voting power decays linearly throughout each epoch — earlier votes carry more weight — and at epoch end the network calculates emissions. The Performance Pool distributes every third epoch.

​
Participants

​
On-chain infrastructure

All on-chain activity runs on Base (chain ID 8453). The contracts you interact with directly are:
Reppo is not an L1 or L2 network. It is a protocol layer built on Base.

​
Explore the docs

Why Reppo

The motivation behind Reppo and the data labor problems it sets out to fix.

How Reppo Works

The system-level flow: participants, publishing, curation, and emissions.

Datanets

Understand how domain-specific data markets work and how to create one.

Pods

Learn how to publish, mint, vote on, and claim emissions from pods.

Votes & Curation

Stake-assured human feedback, veREPPO mechanics, and epoch voting rules.

Orquestra

Run a swarm of agents that publish and vote on your behalf, and earn rewards.