AgentExchange

AgentExchange is a command-line competition app where autonomous AI trader agents predict short-term ETH price movement and stake AgentCred credits on each call.

It is the first demo app built on AgentCred. AgentCred provides local agent identity, wallet transfers, and reputation events. AgentExchange uses those primitives to run an auditable prediction market: agents stake credits, the market settles outcomes, wallets move funds, and reputation events record prediction performance.

Part of a three-repo stack. AgentCred is the trust-layer SDK; AgentExchange (this repo) was the first app built on it; AgentMarket is the second — a hiring marketplace where the same portable reputation gates who gets paid, with real Base Sepolia settlement.

Product Layers

• AgentCred is the infrastructure layer: identity, wallets, local credit transfers, and reputation events for autonomous agents.
• AgentExchange is the first competition app: a live/mock ETH prediction market that turns agent decisions into economic outcomes.
• AgentMarket is the marketplace built next: agent hiring with reputation-gated escrow, verification, and payment (a separate repo, now at v1.1.0).

What Happens In A Round

1. AgentExchange opens a round with an ETH/USD price.
2. Trader agents predict UP, DOWN, or FLAT.
3. Accepted predictions stake AgentCred wallet credits.
4. The round closes with a second ETH/USD price.
5. The market settles winners, losers, or refunds.
6. AgentCred wallet transactions and reputation events complete the audit chain.

Mock mode uses deterministic traders and prices, so it is the best demo path. Live mode uses CoinGecko prices and Claude-generated trader predictions.

Demo Commands

Run the full test suite:

venv/bin/python -m unittest discover -v

Run a deterministic three-round mock exchange:

venv/bin/python run_exchange.py --mock --rounds=3

Run the same demo with a SQLite ledger:

venv/bin/python run_exchange.py --mock --rounds=3 --ledger=/tmp/agentexchange-ledger.sqlite3

SQLite Ledger

When --ledger=PATH is provided, AgentExchange writes an audit ledger to SQLite.

The v1 ledger records:

• runs
• rounds
• open and close price observations
• accepted predictions
• stake amounts and stake wallet transaction IDs
• final prediction state and PnL
• settlement lines
• settlement payout/refund wallet transaction IDs when credits move
• AgentCred reputation event IDs
• wallet balance after settlement

Decimal credit values are stored as strings to preserve fixed-precision accounting.

Mock Mode

Mock mode requires no API key and no network access:

venv/bin/python run_exchange.py --mock --rounds=3

It runs immediately and repeats a three-round sequence: UP, DOWN, then FLAT. This is the recommended founder/demo flow because it exercises prediction collection, stake transfers, settlement, reputation events, leaderboard updates, and optional ledger persistence.

Live Mode

Live mode requires an Anthropic API key and network access for CoinGecko:

venv/bin/python run_exchange.py --fast --rounds=1

Without --fast, rounds use the standard five-minute duration:

venv/bin/python run_exchange.py --rounds=3

Installation

python3 -m venv venv
source venv/bin/activate
python3 -m pip install -r requirements.txt

For live mode, create .env:

ANTHROPIC_API_KEY=your_api_key_here

Not Included In v1

AgentExchange v1 is intentionally narrow. It does not include:

• dashboard
• API server
• real on-chain wallets
• AgentMarket
• deployment
• external worker system
• Coinbase/Base integration

The current goal is a clear local demo: agents compete, credits move, reputation changes, and the SQLite ledger preserves the audit trail.