Tempo Agent Treaty

Agentic OTC block trading for illiquid tokens. AI agents negotiate large block trades P2P using the Vellum negotiation engine, settling via the Machine Payments Protocol (MPP) on Tempo chain.

Quick Start

# Backend
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"

# Copy and fill in your Tempo testnet keys
cp .env.example .env

# Start (mock mode works without keys)
uvicorn backend.app:app --reload --port 8000

# Frontend (separate terminal)
cd frontend && npm install && npm run dev

Architecture

Negotiation (Vellum SDK)          Settlement (MPP on Tempo)
========================          ========================

Seller Agent                      Buyer Agent
    |                                 |
    |  propose price=1.05             |
    |  ─────────────────────►         |
    |                                 |
    |         ◄──── counter price=1.02|
    |                                 |
    |  (PolicyEngine evaluates        |
    |   cross-field concessions)      |
    |                                 |
    |  accept price=1.03              |
    |  ─────────────────────►         |
    |         ◄──── accept            |
    |                                 |
    |  ═══ ALL 12 FIELDS AGREED ═══   |
    |                                 |
    |  SHA256(agreed_terms) ──────►   |
    |                           402 Challenge
    |                          {pay \$103K USDC}
    |                                 |
    |                      auto-pay via pympp
    |                                 |
    |                      ◄── Receipt + tx_hash
    |                                 |
    ╰── SETTLED ON TEMPO CHAIN ──────╯

Vellum SDK — General purpose negotiation engine (Python + React)

• 12 negotiable fields across 4 sections (economics, execution, trust, market reference)
• Cross-field concession logic (wider slippage if more tranches offered)
• Field-level state machines, consensus engine, proposal manager

Policy Engine — Hard guardrails per agent

• Deterministic ACCEPT/REJECT/COUNTER decisions
• REFER_TO_LLM for ambiguous zones (auto-counter in full-auto mode)

MPP Settlement — Machine Payments Protocol on Tempo chain

• Buyer pays seller via HTTP 402 challenge/response (pympp SDK)
• Content hash of agreed terms embedded in payment memo
• Falls back to mock settlement when no keys configured (MPP_MODE=mock)

How it works

Two agents with competing objectives (best price vs best fill) negotiate a block trade across 12 parameters at once: price, quantity, slippage tolerance, settlement window, tranche count, escrow terms, penalty rate, expiry, oracle source, and TWAP reference. The policy engine enforces cross-field concession logic. When both sides agree, the trade settles atomically via MPP on Tempo chain. Agents also query market data via MPP-gated oracle endpoints to compute their BATNA (Best Alternative To Negotiated Agreement) against AMM execution.

Configuration

Variable	Description	Default
MPP_MODE	live for real chain, mock for demo	mock
TEMPO_SELLER_PRIVATE_KEY	Seller wallet (testnet)	—
TEMPO_BUYER_PRIVATE_KEY	Buyer wallet (testnet)	—
TEMPO_CHAIN_ID	Tempo chain ID	42429
TEMPO_RPC_URL	Tempo RPC endpoint	testnet

Tests

python -m pytest backend/tests/ -v  # 74 tests, ~0.3s