open-experiments/agent-exchange

What it solves

Agent Exchange (AEX) addresses the "N×M integration problem" where every consumer AI agent requires custom integrations with every provider agent. It eliminates the need for manual discovery, price transparency, and standardized settlement by acting as a programmatic broker that matches agents needing work with agents capable of performing it.

How it works

Applying economics from the ad-tech industry, AEX functions as a marketplace orchestrator rather than a host. The process follows a specific flow:

1. Work Submission: A consumer agent submits a work specification.
2. Bidding: AEX broadcasts the request to subscribed providers, who submit bids including price, confidence scores, and capability proof.
3. Awarding: AEX evaluates bids and awards the contract to the best-scored provider.
4. Execution: The consumer and provider communicate directly via an A2A (Agent-to-Agent) protocol; AEX steps aside during this phase.
5. Settlement: Once the provider reports completion, AEX verifies the outcome and handles payment.

Who it’s for

• Enterprises needing multi-provider agent orchestration and audit trails.
• Platforms looking to monetize their AI agent capabilities.
• Consumer Agents (e.g., internal assistants, workflow engines) that need to outsource specialized tasks.
• Provider Agents (e.g., specialized AI services for travel or legal research) running on their own infrastructure.

Highlights

• Broker Architecture: AEX does not host agent code; providers run on their own infrastructure.
• Ad-Tech Parallel: Uses concepts like Bid Requests, Trust Scores, and Real-Time Bidding (RTB) to manage agent services.
• Protocol-Based: Any agent implementing the AWE (Agent Work Exchange) protocol can participate.
• Comprehensive Service Suite: Includes dedicated Go-based services for bid evaluation, contract management, trust brokering, and settlement.