Autonomerce

Inspiration and problem

AI agents can already perform valuable digital work, but most agent owners still operate the commercial layer manually. They must decide what the agent can sell, package the capability, find an interested buyer, negotiate safe terms, collect payment, route the task, decide whether delivery passed, and preserve evidence. That repeated integration work prevents a useful agent from becoming a repeatable small business.

Autonomerce asks a practical entrepreneurship question: what if connecting an agent to a sales department were as repeatable as connecting it to a tool?

What Autonomerce does

Autonomerce is a policy-controlled commercial operating layer for callable AI agents. Its portable component, OfferRail, converts a declared A2A, MCP, OpenAPI, or custom capability into a machine-readable service SKU. An owner defines the commercial and wallet envelope once. Inside that envelope, Autonomerce can match an explicitly opted-in buyer need, create a proposal, handle a bounded counteroffer, request USDC settlement, route paid work to the seller agent, validate the returned artifact separately, and publish a redacted receipt.

The core design rule is that adaptive reasoning is not financial authority. Gemini may recommend product framing, relevance, or a negotiation action, but deterministic code controls price, scope, capacity, buyer eligibility, token, chain, payer, payee, amount, idempotency, payment confirmation, and delivery acceptance. Payment never automatically means the work succeeded.

How Gemini is operational

Gemini is the adaptive productization and sales-reasoning layer, not decorative copywriting. The deployed private Cloud Run API uses gemini-2.5-flash to transform a declared source-verification capability into a structured SKU with an outcome, price, latency, capacity, and acceptance criteria. The public evidence records the requested model, API revision, timestamp, request latency, structured result, and resulting SKU. The owner policy then clamps the output so the model cannot expand its authority.

Removing Gemini would remove adaptive productization and sales reasoning. Removing the deterministic policy would make the system unsafe; therefore both layers are necessary.

How Circle is central

Circle is designed as the settlement rail bound to the accepted commercial contract, not as an unrelated checkout button. The payment lane validates the canonical USDC asset, exact amount, network, payer, payee, wallet allowlists, per-payment cap, cumulative cap, and payment count before execution. Durable SQLite state binds one accepted proposal to one idempotency contract. Ambiguous outcomes stop for reconciliation instead of being blindly retried.

The current public proof includes one founder-owned Circle Agent Wallet transfer of exactly 0.10 USDC on Arc testnet. Circle history, before-and-after balances, and independent Arc JSON-RPC verification all match the same transaction. A durable replay returned the same proposal, payment, and transaction identifiers without a duplicate transfer. Two earlier attempts are also disclosed: one failed before Circle code ran because Node was unavailable, and one was rejected before submission because Circle required a UUID-form idempotency key.

This is testnet integration evidence, not customer revenue. The deployed Gemini order still uses offline payment, so the Gemini and Circle traces are not yet one deployed end-to-end customer order. The sponsor wording requires a real, verifiable USDC transaction without explicitly confirming that testnet alone is sufficient; this submission therefore does not treat the testnet receipt as a final eligibility determination.

Human and AI responsibilities

The owner remains responsible for account terms, authentication, wallet funding, standing policy, emergency stop, customer consent, evidence publication, legal attestations, and final Devpost submission. Gemini and the software operate only inside the approved envelope. No model can change the destination wallet, raise the cap, count a testnet transfer as revenue, publish private customer content, or accept its own seller output.

Economic opportunity and business model

Autonomerce is intended for builders who own a useful agent but do not want to build a bespoke sales and settlement stack for every capability. The proposed business model is a fee or revenue share on successfully settled and accepted agent services, with optional paid onboarding. The first wedge is a source-verification and evidence-pack service because its inputs, outputs, and acceptance contract are machine-checkable.

The repository defines seller activation, external customers, paid tasks, refunds, variable costs, and gross margin, while explicitly excluding synthetic, testnet, self, founder, affiliate, reimbursed, and circular activity from customer revenue. No external customer, revenue, margin, repeat-purchase, or job-creation number is claimed yet. Those are the next business-validation steps, not values to infer from the technical demo.

Build and current result

The public Next.js application connects through a server-side BFF to a private IAM-protected FastAPI service on Cloud Run. The repository includes typed contracts, exact-decimal USDC handling, bounded negotiation, durable payment state, independent receipt verification, reconciliation, fulfillment validation, redaction, threat modeling, public-evidence schemas, and a credential-free deterministic demo.

The clean release preflight passes 460 Python tests and 47 web tests, reproduces the offline demo byte-for-byte, scans the public tree for likely secrets, typechecks the web application, and builds the production bundle. These results prove the tested software and the specific deployed/testnet integrations described above. They do not prove production readiness, external demand, revenue, or first place.

Autonomerce’s north star is simple: give an existing AI agent a policy-bounded path from capability to offer, settlement, independently judged delivery, and auditable receipt.

Built With

• a2a
• arc-testnet
• circle-agent-wallets
• docker
• fastapi
• github-actions
• google-cloud-run
• google-gemini-api
• mcp
• next.js
• openapi
• python
• sqlite
• typescript
• usdc
• vertex-ai