Pundi AI is a protocol for collecting, labeling, and verifying AI training data onchain, using PUNDIAI tokens and reputation scores to incentivize data quality.
Users earn PUNDIAI token incentives by tagging and verifying data through a Tag-to-Earn system, reducing reliance on centralized data labeling.
Pundi AI’s integration with ElizaOS connects its datasets to multi-agent systems, enabling agents to train and coordinate using onchain datasets verified by human contributors.
Partnerships with Twin3, Hooked, Swarm, and SirenAI extend Pundi AI’s reach into decentralized identity, education, and agent design.
In April 2025, Pundi AI joined the NVIDIA Inception program, gaining access to discounted hardware, cloud credits, and training through the NVIDIA Deep Learning Institute.
Primer
Pundi AI (PUNDIAI) is a decentralized data infrastructure platform that incentivizes sourcing and labeling high-quality datasets for AI. Through its Tag-to-Earn model, users verify and annotate data to earn PUNDIAI token rewards, addressing a key bottleneck in AI development: the need for accurate, human-labeled inputs. While access to clean data is essential, many AI models rely on a centralized provider of data labels to tag, classify, and rank examples. These labels are essential for guiding model behavior, but producing them can be costly, labor-intensive, and, in some cases, unethical. Pundi AI aims to solve the bottleneck through matching third-party datasets with a decentralized network of data providers who earn rewards for labelling data.
The platform is built on Pundi AIFX, a Layer-1 network rebranded from Function X’s f(x) Core. Pundi AI spans three layers: an omnichannel data layer (AIFX), integrated L1/L2s, and an application layer, which supports cross-chain data movement and EVM compatibility.
Developers can train and deploy AI agents via the Pundi AI MM Agent Launchpad, then launch tokens using bonding curves. A market-making agent (Pundi AI MM) provides onchain liquidity automation by placing and managing orders based on parameters and real-time market data. PURSE+ is a browser plugin that lets users earn rewards by tagging and analyzing social media data to support community-driven AI development. It is part of Pundi’s broader AI initiative, which focuses on enhancing AI capabilities through community-driven efforts.
PUNDIAI is the native token for payments, incentives, and protocol governance on the Pundi AI network. vePUNDIAI is a vote-escrowed (locked for a set period) version of the PUNDIAI token that grants holders governance rights over PUNDIAI emissions and protocol decisions. Voting power increases based on the amount locked and the lock duration. Holders can direct PUNDIAI incentives to different products on Pundi AIFX. External protocols or users can bribe vePUNDIAI holders by offering rewards in exchange for their governance votes, often to incentivize governance decisions and liquidity support.
According to an August 2024 Weka Survey, AI model performance today is challenged by data quality more than architecture or compute in many current systems. While model architectures and deployment capacity have advanced rapidly, data quality has not kept pace. Datasets that are publicly auditable and accurately labeled are limited in availability. Datasets are typically controlled by private entities and costly to annotate. For autonomous agents, this lack of quality data reduces agents’ ability to adapt to context-specific tasks or environments. With agents beginning to operate autonomously in user-facing applications, data quality is crucial for safe decision-making.
This gap is worsened because AI development today is controlled by a few firms with opaque data labelling standards and limited contributor compensation. A handful of firms control infrastructure, hoard data, and set opaque rules. Contributors are rarely compensated, and datasets are often biased, outdated, or mislabeled. Models trained using low-quality labels can rely on false patterns, leading to hallucinations, biased outputs, and unstable behavior. This degrades agent reliability, increases the risk of harmful decisions, and undermines safety in real-world applications.
Pundi AI plans to address this with their upcoming decentralized data platform, where data sourcing, labeling, and point of origin are transparent and verifiable onchain. Contributors earn PUNDIAI token rewards, communities govern curation, and agents are trained on traceable, high-integrity datasets.
How Pundi AI Creates Better Agents
Reinforcement Learning from Human Feedback (RLHF) is a machine learning technique that uses human input to guide models toward more efficient self-learning. RLHF aligns outputs with human goals and values by training models to make reward-maximizing decisions. Human feedback is incorporated into the reward function when explicit goals are hard to define, allowing models to learn desired behaviors from preference comparisons.
Pundi AI improves the RLHF process using an incentive-driven platform to collect diverse data labels. The data can be used to fine-tune models and agents with human values and preferences. Pundi AI provides transparency and fair compensation in the data labeling process, which the team hopes will lead to more trustworthy and consistent training data.
Additionally, each step in the data lifecycle (submission, labeling, and verification) is recorded onchain, with each labeling step stored onchain, enabling full auditability. Contributors are weighted by reputation to filter out inaccurate labels across tasks. This improves data integrity and reduces mislabeled inputs during training, which is important for agents operating in environments like healthcare or finance, where model errors can result in significant real-world consequences. Pundi supports continuous improvement after deployment by letting developers retrain models with real user interactions to improve accuracy and response relevance.
Beyond data curation, Pundi includes a launchpad (the Pundi AI MM Agent Launchpad) for users to deploy agents as smart contracts. AI agents come with a token issued using a bonding curve that sets token prices algorithmically based on circulating supply. These curves provide built-in liquidity, with token holders voting on parameter changes, retraining cycles, or access to datasets.
ElizaOS Partnership
Pundi announced their integration with ElizaOS in June, bringing decentralized, human-labeled data directly into agent development workflows. Developers can now access and ingest labeled onchain datasets for training, making training agents on transparent, verifiable data easier.
ElizaOS provides a framework for deploying multi-agent systems with shared state and identity. Built on Solana, it supports real-time coordination at scale. Through Pundi, ElizaOS agents can incorporate human feedback loops and onchain, human-labelled datasets into model development.
Future upgrades include direct buying, selling, and verification of datasets onchain, preprocessing workflows (cleaning and structuring raw data before training), and model-centric processing (MCP). Agents may request or generate data autonomously using Pundi’s infrastructure in the long term. These capabilities will connect data labeling, model tuning, and post-deployment learning through one verifiable pipeline.
Other Partnerships
Several of Pundi AI’s recent collaborations are focused on improving AI agents with better data quality, transparency, and deployment infrastructure:
Twin3 AI (March 7, 2025): Twin3 AI uses Soulbound Tokens (SBTs) to provide users with digital identities. Pundi AI will integrate its AI MM Agent (an automated, onchain market-making bot) with Twin3’s identity framework to give agents their own verified digital identities.
Hooked Protocol (April 22, 2025): Hooked Protocol is a Web3 social learning platform. As an educational partner, Pundi AI launched an introductory module on Hooked covering decentralized approaches to AI development and governance.
Swarm Network (May 5, 2025): Swarm Network enables decentralized multi-agent coordination and onchain data validation. Through its integration with Pundi AI, developers gain access to human-verified datasets critical for training agents that operate autonomously on blockchains. The partnership reduces misfires in autonomous tasks by providing agents with labeled training data.
SirenAI (May 7, 2025): SirenAI agents switch between structured data parsing (extracting and organizing information from raw inputs) and user-facing language generation (translating data into readable responses for end users). Just like Swarm Network’s integration with Pundi AI, developers gain access to human-verified datasets critical for training agents that operate autonomously on blockchains.
NVIDIA Inception Program
In April 2025, Pundi AI joinedNVIDIA Inception, a program that supports early-stage companies developing AI and machine learning. As part of the program, Pundi AI will gain access to resources, including discounted hardware, cloud credits, and technical training through the NVIDIA Deep Learning Institute.
The collaboration opens access to NVIDIA’s network of AI researchers and startups, which could support Pundi AI’s efforts to develop and distribute labelled data. As the project expands its data platform and marketplace, NVIDIA’s high-performance infrastructure and industry expertise could help lower operational costs and accelerate delivery cycles.
Closing Summary
Pundi AI is building a decentralized data infrastructure to solve one of AI’s core challenges: access to high-quality, labeled data. Its Tag-to-Earn system compensates contributors for labeling and verifying data, creating a transparent pipeline for supervised learning and RLHF processes. Each step in the data lifecycle is recorded onchain, and contributors are ranked by reputation to filter out low-quality inputs. This structure improves the reliability of AI agents across all environments, with the greatest impact in high-stakes domains like healthcare and finance due to the severe consequences of errors.
The platform combines data curation with deployment tools. Developers can launch agents using bonding curves, manage them through tokenized governance, and incorporate real-world feedback into training cycles. Recent partnerships, including integrations with ElizaOS, Swarm Network, Twin3 AI, and SirenAI, expand access to verifiable data and support decentralized agent coordination, identity verification, and educational outreach. These developments reflect Pundi AI’s broader goal: to offer an open alternative to proprietary pipelines that lack transparency and contributor incentives.
This report was commissioned by PUNDI AI Finance. All content was produced independently by the author(s) and does not necessarily reflect the opinions of Messari, Inc. or the organization that requested the report. The commissioning organization may have input on the content of the report, but Messari maintains editorial control over the final report to retain data accuracy and objectivity. Author(s) may hold cryptocurrencies named in this report. This report is meant for informational purposes only. It is not meant to serve as investment advice. You should conduct your own research and consult an independent financial, tax, or legal advisor before making any investment decisions. Past performance of any asset is not indicative of future results. Please see our Terms of Service for more information.
No part of this report may be (a) copied, photocopied, duplicated in any form by any means or (b) redistributed without the prior written consent of Messari®.
Sam was previously a product manager at ether.fi, as well as LifeMed AI. He graduated from Ohio State University, where he founded the Ohio State Blockchain Club. He is now an Enterprise Research Analyst at Messari focused on finding the next best applications and protocols in internet finance.
Sam was previously a product manager at ether.fi, as well as LifeMed AI. He graduated from Ohio State University, where he founded the Ohio State Blockchain Club. He is now an Enterprise Research Analyst at Messari focused on finding the next best applications and protocols in internet finance.
