AEP – On-chain marketplace where CrewAI agents can hire each other and get paid (Base Mainnet)

[TomsonTrader]

Hey — built something you might find interesting if you work with autonomous agents.

AEP (Autonomous Economy Protocol) is a fully on-chain marketplace on Base where AI agents
register, negotiate prices, and execute escrow deals — no humans involved.

Ships with a LangChain toolkit (11 ready-made tools) so your agents can participate in
seconds:

npm install autonomous-economy-sdk

GitHub: https://github.com/TomsonTrader/autonomous-economy-protocol
Dashboard: https://autonomous-economy-protocol-1.vercel.app

Happy to answer questions.

[aniruddhaadak80]

From my point of view, the interesting question here is not only whether agents can pay each other, but how you constrain the decision path before an escrow gets opened.

If this is meant to plug into AutoGen style loops, I would want every economic action to be quote first, commit second, and fully idempotent under retries. Otherwise one planning mistake or network retry turns into duplicated settlement very quickly.

I would also be curious how you handle budget caps, dispute resolution, and partial task completion. Those three pieces usually decide whether an agent marketplace feels like infrastructure or just a demo.

[TomsonTrader]

You're pointing at exactly the gap between web3 infrastructure and production agent systems. The protocol has separate states and dispute resolution, but you're right that idempotency under retries isn't solved yet .it's on the roadmap as a quote-id/nonce layer above the NegotiationEngine.

The LangChain toolkit wraps each action as a deterministic function call, not free form intent. propose_deal(need_id, offer_id, price) is as constrained as an API call.

The reconciliation problem is the hardest one .would love to hear how you'd model it. The current pattern is getProposal(id) polling but you're right it's not first-class.

[LortuArte]

@TomsonTrader You hit the nail on the head. Neither L1 nor standard L2 rollups can survive high-frequency agent micro-transactions (hiring, API calls, data sharing) without bleeding margins on gas and breaking the AI's cognitive loop with block latency.

The exact architecture you are looking for is an off-chain state channel (L3) for the hot-path, settling the batched receipts on Base (L2) later.

I built AEGIS Protocol: an off-chain M2M settlement engine in Python. It acts as an L3 clearinghouse, clearing internal micro-transactions between agents at 0.005ms with zero gas fees.

I've provisioned an isolated test node on the production ledger for the Auto Company OS research team and pre-funded it with $10.00 USD so you can benchmark the sub-millisecond latency against your current Base setup.

Drop me a DM here lortuarte765@gmail.com and I'll hand over the API keys.