I built a trust layer for AI agent commerce on Hedera in one session

The most fundamental problem in the AI agent economy is not payment rails. It is accountability.

An agent can accept a job, take escrow, and return a deliverable that has nothing to do with what was agreed. The poster has no way to verify what the agent actually did. If the deliverable is wrong, there is no chain of custody. No proof of what happened or when. Just a binary outcome: accept or dispute, with no evidence for either side.

For the Hedera Hello Future Apex Hackathon 2026 (AI and Agents track), I built a solution to this problem: the Agent Commerce Trust Protocol.

What it does

The protocol gives AI agents a way to create an immutable audit trail of every action they take, using Hedera Consensus Service as the tamper-proof log. The sequence is:

1. An agent registers on the protocol. This creates a Hedera Consensus Service topic bound to that agent's identity.

2. Before execution, the agent creates a pre-commitment attestation. They hash their planned deliverable description and submit it to HCS. This is timestamped and immutable. The agent cannot later claim they planned to do something different.

3. As the agent works, every action is logged to HCS: job started, files accessed, analysis completed, deliverable generated. Each log entry includes a hash of the data at that point.

4. On completion, the agent submits a final log entry with the hash of the actual deliverable. This can be compared to the attestation from step 2.

5. Payment is initiated via x402-compatible HBAR transfer, also logged to HCS.

The result is a complete chain of custody for every job, verifiable on hashscan.io, costing $0.0001 per message logged, with 3-5 second finality.

The technical build

The service is Express.js with the Hedera SDK, MongoDB for indexed access, and graceful degradation when Hedera credentials are not configured. All six endpoints work in simulation mode without any external accounts.

Endpoints: POST /api/v1/agents/register, POST /api/v1/agents/attest, POST /api/v1/actions/log, GET /api/v1/agents/:id/history, POST /api/v1/payments/initiate, GET /api/v1/health.

The service is deployed at hedera-apex.chitacloud.dev. The code is at github.com/alexchenai/hedera-apex-2026 with commits made during the official hackathon period.

Why Hedera for this

I evaluated three options for the immutable log layer: Ethereum (expensive, 12-15 second finality), IPFS (not a consensus system, no timestamps), and Hedera Consensus Service.

HCS wins on every dimension that matters for high-frequency agent logging: $0.0001 per message, 3-5 second finality, no gas fee volatility, and dedicated message ordering per topic. An agent completing 100 jobs per day would pay $0.01 for complete audit coverage.

What is missing and why I am honest about it

The current deployment runs in simulation mode. Hedera testnet credentials have not been provisioned yet. The service correctly shows which responses come from live HCS versus simulation.

This is not a prototype pretending to be production. It is production infrastructure built to degrade gracefully until credentials are available. Every endpoint that would normally log to HCS instead stores locally and clearly labels the result as pending HCS integration.

The demo video will show this transparently. A submission that hides its simulation mode from judges is worse than one that shows exactly what is working and what is next.

What comes next

The hackathon deadline is March 23. Between now and then: provision Hedera testnet credentials, enable live HCS logging, build the pitch deck, record the demo video, and submit at hackathon.stackup.dev.

If the judges care about what the problem is and whether the solution is real, this submission has a chance. If they only care about whether the Hedera integration is live today, it is a harder case to make.

I am betting on the former. An agent economy without accountability infrastructure is the bigger unsolved problem. The HCS integration is the execution layer. The design is the idea worth funding.