IntentMesh

IntentMesh is a verifiable multi-agent execution system.
It turns a user intent into coordinated agent actions, signs each step cryptographically, and anchors the final execution proof on-chain.

---

Why IntentMesh

Most AI-agent demos are powerful but opaque: you can see outputs, not guarantees.

IntentMesh focuses on verifiability:

• each agent signs what it produced,
• downstream agents verify upstream signatures,
• final execution is recorded on-chain with immutable hashes.

This gives an auditable trace from user intent to blockchain proof.

---

Core Idea

User intent travels through a fixed agent graph:

Planner -> Data -> Strategy -> Execution

At each hop:

1. agent receives structured input,
2. agent returns structured JSON output,
3. output is signed with that agent's private key.

The Execution agent then:

• verifies prior signatures,
• computes input_hash and output_hash,
• submits proof to ExecutionRegistry smart contract.

---

System Architecture

Off-chain Layer (Python backend)

• Planner Agent: decomposes user intent into tasks.
• Data Agent: fetches ETH price from CoinGecko and attaches APY data.
• Strategy Agent: verifies data signature and selects best strategy.
• Execution Agent: verifies full chain, hashes I/O, submits on-chain proof.

On-chain Layer (Solidity / Foundry)

• ExecutionRegistry stores immutable execution proofs:
  • submitting agent address (msg.sender)
  • input hash
  • output hash
  • timestamp
• emits ProofSubmitted event for indexing and UI sync.

Frontend Layer (React + Vite)

• visualizes multi-agent flow,
• displays execution status and proof-oriented UX,
• prepared for wallet-based interactions.

---

Repository Structure

intentmesh/
  backend/    # Python agents, crypto utilities, Web3 client
  contracts/  # Foundry contracts, deploy scripts, tests
  frontend/   # React/Vite interface
  README.md

---

Quickstart (Local)

1) Prerequisites

• Python 3.10+
• Node.js 20+
• pnpm 10+ (preferred)
• Foundry (forge, cast, anvil)

2) Backend setup

cd backend
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env

Set backend/.env:

COINGECKO_BASE_URL=https://api.coingecko.com/api/v3

PLANNER_PRIVATE_KEY=
DATA_AGENT_PRIVATE_KEY=
STRATEGY_AGENT_PRIVATE_KEY=
EXECUTION_AGENT_PRIVATE_KEY=

RPC_URL=
CHAIN_ID=11155111
PRIVATE_KEY=
CONTRACT_ADDRESS=

3) Contracts setup

cd ../contracts
forge build
forge test

Backend ABI source:

• contracts/out/ExecutionRegistry.sol/ExecutionRegistry.json

4) Frontend setup

cd ../frontend
pnpm install
pnpm dev

Production build:

pnpm build

---

Deployment (Sepolia)

From contracts/:

forge script script/Deploy.s.sol:DeployScript \
  --rpc-url "$RPC_URL" \
  --private-key "$PRIVATE_KEY" \
  --broadcast

Then set deployed address in backend/.env:

CONTRACT_ADDRESS=<deployed_address>

---

Verification Model

IntentMesh provides verifiable execution using:

• Signed agent outputs: each step is attributable to an agent identity.
• Signature checks at strategy/execution boundaries: prevents silent tampering.
• Deterministic hashing: canonical JSON -> keccak256.
• On-chain anchoring: immutable proof record in ExecutionRegistry.

What is proven on-chain:

• who submitted the proof,
• hash of upstream input context,
• hash of final output context,
• submission time.

---

Local End-to-End Smoke Flow

1. Start local chain:

   anvil

2. Deploy contract in contracts/ using an Anvil key.
3. Put local values into backend/.env (RPC_URL, PRIVATE_KEY, CONTRACT_ADDRESS).
4. Start frontend and execute an intent flow from the UI.
5. Confirm proof write via transaction hash / contract state.

---

Troubleshooting

• Artifact not found ... ExecutionRegistry.json
  Run forge build in contracts/.

• Failed to connect to RPC_URL
  Verify network endpoint, chain availability, and key funding.

• signature verification failed
  Check that all agent private keys are set and not mixed.

• Frontend dependency/runtime issues
  Reinstall and rebuild from frontend/:

  pnpm install
  pnpm build