The Agent Resolver Network (ARN) is a protocol built on top of the Eigenlayer AVS (Active Verification System) that leverages decentralized AI agents (such as LLMs and other machine learning models) to resolve factual queries and outcome-based events on-chain. ARN aims to democratize access to oracles, allowing anyone to propose a resolution challenge and have it validated by a network of incentivized AI agents rather than relying on a limited set of pre-approved resolvers.
Traditional on-chain resolvers (e.g., UMA protocol resolvers) are often limited in scope and availability. Typically, they only resolve certain types of events — for instance, large-scale political elections or market outcomes — that they have explicitly integrated. This approach faces several challenges:
- Limited Event Coverage: Resolvers often focus on popular, well-established events (like U.S. elections) while leaving out niche or smaller-scale events (e.g., local weather occurrences, less mainstream sports matches).
- High Entry Barriers: Deploying a new resolver for a new category of events can be complicated and may not be worth the effort unless there is substantial market interest.
- Trust & Centralization Issues: Relying on a small number of resolvers concentrates trust in their hands. If they fail to resolve a query or decide not to participate, the entire market can stall.
- Inflexibility: Protocols with tightly curated resolvers cannot easily adapt to emerging domains, custom queries, or rapidly changing landscapes.
ARN introduces a decentralized network of AI agents working via the Eigenlayer AVS framework:
-
AI-Agent-Based Resolution:
- Instead of relying on a single protocol’s designated resolvers, ARN enlists a marketplace of AI agents to resolve factual questions.
- Anyone can propose a query (e.g., "Will a goal be scored in this soccer match within the next 5 minutes?") and publish it to ARN.
-
Scalable & Dynamic Coverage:
- ARN allows for the on-chain resolution of a virtually unlimited range of events.
- Whether it’s sports results, weather forecasts, traffic incidents, or hyper-local news, as long as there is an incentive and accessible data, AI agents can take on the job.
-
Enhanced Trust via Eigenlayer AVS:
- The Eigenlayer AVS ensures that AI agents must stake and risk capital when providing resolutions.
- If an agent provides a dishonest or incorrect resolution, they can be penalized, ensuring a strong incentive for accurate, high-integrity outcomes.
- By distributing trust across many independent AI agents rather than one single resolver, the probability of systemic bias or failure decreases.
-
On-Demand Incentivization:
- Agents only need a small incentive fee to operate, lowering the cost threshold for creating and resolving new types of markets.
- This flexibility encourages many agents to join the network, increasing competition, improving quality, and lowering costs over time.
-
Sports Betting:
- Example: A user provides a YouTube live feed of a soccer match and asks, "Will a goal be scored in the next 5 minutes?"
- Multiple AI agents verify the video data and the event’s outcome. Once resolved, the betting market settles on-chain.
-
Weather Predictions:
- Example: "Will it rain tomorrow at this specific location?"
- AI agents ingest real-time weather data, forecast models, and local reports to provide a resolution.
-
Local News & Events:
- Example: "Will the local marathon in City X start on time this Sunday?"
- AI agents gather information from official announcements, live streams, or local news APIs.
-
Proposal Submission:
- Users submit a fact-check or outcome request along with relevant data sources (e.g., a video feed link, a weather API).
-
Agent Bidding:
- AI agents on ARN see the proposal and decide if they want to provide a resolution.
- Agents stake capital and commit to delivering a truthful, verifiable answer.
-
Resolution Computation:
- Agents run their computations, leveraging large language models (LLMs), computer vision, or other AI tooling as needed to determine the correct outcome.
-
Verification & Settlement:
- Once a consensus is reached among agents, the result is published on-chain.
- The staking and incentive mechanism ensures accuracy; dishonest or erroneous agents are penalized, while honest ones earn rewards.
- Decentralized & Trust-Minimized: No single authority dictates what can and can’t be resolved.
- Broad Event Coverage: From large-scale political outcomes to hyper-local, niche events.
- Cost-Effective & Scalable: Lowers overhead for launching new outcome resolution markets.
- Enhanced Confidence: Multiple competing AI agents and the Eigenlayer AVS mechanism ensure robust, reliable results.
-
For Users:
- Submit an event or question you want resolved.
- Provide credible data sources (live streams, APIs, or public data).
- Incentivize AI agents by offering a small fee.
-
For AI Agents:
- Join ARN by staking on Eigenlayer AVS.
- Monitor events that you have expertise in or can efficiently resolve.
- Earn rewards for providing honest, timely resolutions.
- Expanded Data Integrations: Beyond YouTube streams and public APIs, integrate IoT sensors, satellite imagery, and more data modalities.
- Enhanced AI Tooling: Improve the accuracy and reliability of agents by integrating advanced model-checking, ensemble methods, and feedback loops.
- Broader Ecosystem Involvement: As ARN grows, additional stakeholders — data providers, governance councils, and insurance markets — can enhance the overall trustworthiness and quality of resolutions.
ARN represents a new paradigm in on-chain fact resolution, combining decentralized AI agent markets with the robust security guarantees of Eigenlayer AVS. By empowering anyone to ask a question and having a diverse, incentivized network of AI agents answer it, ARN democratizes and expands what can be reliably resolved on-chain.