CoinGraph Lets AI Agents Pay for Risk Checks on Cardano

The TOKEN2049 Origins winner combines crypto market assessments with payments through x402 and Masumi. Its public prototype lets agents purchase analysis on Cardano’s test network and receive responses backed by sources and verifiable records.

By SongMarketCap

Updated:

Cardano News - CoinGraph Lets AI Agents Pay for Risk Checks on Cardano

CoinGraph, the overall winner of the TOKEN2049 Origins hackathon from the Cardano track, has built a service for AI agents to consult before initiating financial transactions. It evaluates a proposed action against market conditions and risk rules, returning findings that the requesting application can use before committing funds.

CoinGraph Checks Whether an Order Fits the Market

The system evaluates seven areas: momentum, liquidity, leverage, on-chain activity, supply, contract risk, and market context. Findings include sources and timestamps, while areas without sufficient measurements are marked “unassessed.”

According to its documentation, CoinGraph continuously monitors the 100 largest tokens, including ADA, and provides lighter checks on demand for more than 21,000 others. It combines exchange data with information from CoinGecko, NOWNodes, GoPlus, and DefiLlama.

Ten specialist agents turn those inputs into specific services. Wallet Guard checks transfers and swaps before signing, Due Diligence Analyst produces token assessments, and Treasury Steward checks assets against treasury rules.

Developers can access the functions through a REST API or an MCP server compatible with Claude and Cursor.

Trade Gatekeeper applies the analysis to a proposed order’s size and the caller’s limits. It returns ALLOW, REDUCE, or BLOCK with supporting reasons.

A recorded example shows a request to buy $10,000 worth of a token with a maximum permitted price impact of 0.25%. CoinGraph estimated that the full order would have a 0.41% impact and returned REDUCE, proposing approximately $6,100 instead.

The response connected the proposed reduction to available liquidity. The requesting application retained control over whether to proceed and execute the trade.

x402 and Masumi Handle Cardano Payments

CoinGraph supports two payment routes: purchasing an individual API response through x402 or commissioning an analytical task through Masumi escrow.

With x402, a paid request receives an HTTP 402 Payment Required response containing the payment terms. The agent signs a Cardano transaction and resubmits its request with the payment attached. CoinGraph processes the transaction and releases the assessment with a payment receipt. The flow requires neither a customer account nor an API key.

This implementation operates on Cardano Preprod using test ADA. Mainnet billing and subscriptions remain planned capabilities.

For task-based work, CoinGraph Crypto Analyst is available through Sokosumi. Users describe an assignment in ordinary language, and the service selects the appropriate specialist agent.

Masumi escrow holds the payment while the task proceeds, and the result’s digital fingerprint is recorded on-chain. The documented Preprod example covers funding, result submission, and payment collection after the applicable unlock time and dispute window.

Verifiable Responses and a Public Performance Record

Each stored CoinGraph response receives an identifier and a SHA-256 fingerprint. Recomputing that fingerprint allows users to check whether the content matches the recorded response, alongside the sources used to produce it.

The verification establishes the record’s integrity. The assessment itself remains dependent on the quality of its data and analytical rules.

CoinGraph also includes a Chainlink CRE workflow that recalculates fingerprints and records attestations. It currently runs in the CRE simulator, with deployment to a decentralized oracle network listed as the next step.

A separate public record compares assessments and signals with subsequent price movements over one-hour, 24-hour, and seven-day intervals. Once each interval has elapsed, users can revisit the original response and examine the recorded outcome. The history retains misses alongside successful results, allowing earlier assessments to remain open to scrutiny after market conditions change.