The problem AGORA solves

The Problem

AI agents are arriving — but they have no common language for commerce.

Today, if an autonomous trading agent wants to buy wheat futures from another agent, there is no standard protocol for how to express the intent, agree on a price, or settle atomically. Every team invents custom integrations that only work within their own system. Agent A on one platform cannot trade with Agent B on another. There is no neutral price oracle agents can trust, and no way to exchange value without counterparty risk.

The result: The multi-agent AI economy cannot function at scale.

What AGORA Solves

AGORA is the Universal Commodity Protocol (UCP) — the missing commerce layer for autonomous AI agents.

For AI developers — your agents can publish buy/sell intents in a standardized format, discover counterparties across any platform, and settle trades atomically on Hedera. No custom integrations, no trust assumptions.

For commodity traders — watch 4 autonomous agents trade WHEAT futures in real time, with AI-validated oracle pricing from openBridge and instant HTS atomic swaps. Sub-5-second finality. ~$0.001 per transaction vs $15+ on Ethereum.

For the agentic economy — UCP is to agent commerce what TCP/IP is to data exchange. A universal standard that any agent, on any platform, can implement to participate in autonomous markets.

Key Capabilities

• Standardized intent schema — any agent can post a buy or sell with price, quantity, TTL, and oracle attestation in one JSON message
• AI-validated oracle — openBridge fetches prices from multiple sources, runs AI validation, and anchors attestations to Hedera HCS with cryptographic proof
• Instant matching engine — intents are matched within 1% price tolerance, settling in the same block
• Atomic HTS swap — WHEAT tokens and UCPUSD tokens transfer simultaneously — no escrow, no counterparty risk
• On-chain reputation — every settled trade updates an agent's reputation score, enabling trust in a trustless system
• MetaMask + Hedera — judges can connect a wallet, register an agent on-chain, and post a signed intent in under 60 seconds at agoradapp.netlify.app

Challenges we ran into

Challenges We Ran Into

1. MetaMask ↔ Hedera bridging
MetaMask speaks EVM; Hedera is not EVM-native. Getting wallet_addEthereumChain to correctly add Hedera Testnet (Chain 296) via HashIO RPC, resolve EVM addresses to Hedera account IDs through the Mirror Node, and sign UCP payloads with personal_sign required careful mapping between the two identity systems.

2. Atomic settlement without Solidity
On Ethereum you'd use a smart contract for atomic swaps. On Hedera we had to architect the settlement entirely using HTS scheduled transactions and HCS message sequencing — no Solidity at all. Getting the token transfer to be truly atomic (both sides or neither) required careful use of Hedera's native transaction primitives.

3. Oracle consensus across sources
Aggregating prices from CoinGecko, Chainlink, and Pyth in real time and producing a single AI-validated attestation that agents can trust meant building a lightweight consensus mechanism inside openBridge — detecting outliers, weighting sources by freshness, and anchoring the final result to HCS before any agent can use it.

4. Real-time matching with live oracle drift
Agent buy/sell prices are pegged to oracle offsets (e.g. "buy at oracle + 1%"). As the oracle updates every 10 seconds, previously unmatched intents can suddenly become matchable. Building a matching engine that re-evaluates the entire order book on every oracle tick — without creating race conditions — was a non-trivial concurrency challenge.

5. Instant on-chain identity for demo users
For judges to register an agent on-chain in under 60 seconds, we needed the full flow — MetaMask connect → switch to Hedera Testnet → sign registration payload → broadcast via HashIO — to be frictionless. Most wallet UX for new chains is painful; we iterated the step-by-step UI several times to get it right.

Use of AI tools and agents

How AI Powers AGORA

AGORA is not just built with AI — AI is the operational core of the system. Every trade that settles has passed through multiple AI layers before it executes.

1. openBridge Oracle — AI-Validated Price Feeds

The most critical AI component. Before any agent can post a buy or sell intent, the oracle must attest to the current commodity price.

How it works:

• openBridge fetches WHEAT prices simultaneously from multiple sources (CoinGecko, Chainlink, Pyth)
• An AI validation layer runs outlier detection — if one source is >2% away from the median, it is flagged and down-weighted
• Sources are weighted by freshness — a price updated 3 seconds ago counts more than one updated 45 seconds ago
• The AI produces a single consensus price with a confidence score (e.g. $214.75 | conf: 97%)
• This attestation is anchored to Hedera HCS topic 0.0.7986432 with a cryptographic signature
• No agent can use a price that has not been AI-validated and on-chain attested

Why this matters: Without AI validation, a single manipulated price feed could cause agents to settle trades at wrong prices. The oracle is the trust anchor the entire market depends on.

2. 0G Labs Compute Network — Decentralized AI Inference

The openBridge oracle node is upgraded to use 0G Labs' decentralized compute network for AI inference, rather than a centralized model API.

• Price validation inference runs across the 0G network — no single point of failure
• The AI model weights are verifiable on-chain — any node can confirm the same model produced the attestation
• This makes the oracle itself decentralized: not just the data sources, but the AI reasoning layer too

3. Autonomous Trading Agents — AI-Driven Strategy

The four trading agents (Alice, Bob, Carol, Dave) are fully autonomous AI agents that operate without any human input.

Each agent:

• Continuously monitors the openBridge oracle feed
• Applies its own AI-driven pricing strategy — e.g. Alice always bids oracle + 1%, Bob targets oracle - 0.5% with larger quantities
• Decides autonomously when to post, cancel, or hold an intent based on oracle drift, TTL remaining, and order book state
• Updates its strategy parameters based on past trade outcomes — agents that consistently miss matches adjust their offset thresholds

The agents communicate exclusively through UCP — standardized JSON intents published to HCS. They have no direct connection to each other. The protocol is the only shared language.

4. UCP Matching Engine — AI-Assisted Order Matching

The matching engine does more than simple price comparison.

• On every oracle tick, it re-scores all live intents against the new price — an intent posted at "oracle + 1%" gets a new absolute price computed automatically
• It applies 1% price tolerance matching — if a buyer's max is $218 and a seller's min is $217.30, the AI finds the midpoint ($217.65) and proposes settlement at that price rather than rejecting the match
• Intents are prioritized by reputation score — a 4.8★ agent's intent gets matched before a 3.1★ agent's at the same price, incentivizing good behavior

5. The Full AI Flow — End to End