AI Agents Hackathon

The problem AI agents can't see the physical world. An LLM can reason about anything except where things actually are right now: which satellite is overhead, whether a coordinate is in coverage, what's passing in the next 90 minutes. That data is either unstructured or locked behind enterprise contracts and API keys no autonomous agent can navigate. RealityFrames closes that gap: an authoritative, queryable stream of physical state (orbital first — satellites and coverage via SGP4/TLEs) a machine can consume directly. Agentic AI The customer is the agent, not a person. An agent with skills (search, stream, check-coverage) turns a request like "where is the ISS" into concrete queries against the world model. A resolver sits at the boundary: a request comes in, it either serves the data or demands payment, then resolves. Skills never bill — the resolver does — so another AI can call RealityFrames as a tool and feed verified ground truth into its own world model. Blockchain Settlement is x402: request hits a gated endpoint → 402 Payment Required → agent pays → access granted, all in one programmatic flow. No keys, no signup. Payment is USDC on Base (testnet live; mainnet via Coinbase CDP), with the agent spending from its own Circle wallet. Price maps to product: pay per frame. Why both Payment without agents is just a rail; agents without payment can't transact alone. Together: an agent discovers, pays for, and consumes physical ground truth on its own — no human in the loop.

402Cards lets AI agents buy real-world digital goods with crypto, no cards, no manual checkout. Today, agents can reason and act online, but paying for services still means human wallets, KYC’d accounts, and brittle integrations. 402Cards bridges that gap: an autonomous Circle Agent Stack wallet quotes a Bitrefill Mobile Legends diamond pack, pays via the x402 HTTP payment protocol (USDC on Base), and receives a redemption code, all in one click. A proxy service handles x402 settlement and forwards USDC to Bitrefill, while the agent orchestrates quoting, balance checks, payment approval, and fulfillment polling. Agentic AI: The agent runs end-to-end without hardcoded scripts, it inspects x402 endpoints, validates quotes, manages wallet state, and handles user prompts (e.g. player ID) using Circle’s agent tools and an LLM. Blockchain: Payments use USDC on Base through x402 (HTTP 402 + on-chain settlement), with a single Circle agent wallet as both payer and settlement source, demonstrating machine-to-machine commerce on stablecoins.

Competitive research is slow, expensive, and manual — analysts spend hours Googling, tab-switching, and synthesising scattered results into a brief that's outdated by the time it's written. Scout solves this with an autonomous AI agent that researches a competitor on demand, pays for its own intelligence, and returns a structured brief with receipts — in under 30 seconds. The user enters a company name and a USDC budget. Scout breaks the research into five strategic categories (competitors, recent moves, customer sentiment, pricing, and tech signals), then autonomously pays for each Tavily search query via an x402 nanopayment from a Circle Agent Wallet — $0.01 USDC per query, settled on Base. Every payment produces an on-chain receipt. A hard budget cap ensures the agent never overspends. Once all five searches complete, Nebius Token Factory (Qwen3-30B) synthesises the raw results into a structured competitive brief. The user receives the full brief alongside a spend ledger showing exactly what each insight cost and why the agent decided it was worth buying. Scout demonstrates a new pattern for agentic AI: agents that don't just reason, but pay — autonomously acquiring the data they need, one micropayment at a time, with full accountability on-chain.

Language learning tools are expensive and generic. You pay flat subscriptions for content that may not interest you. This agent finds real native content on topics you actually care about and pays only for what it uses — per lesson, not per month.

HyperFlow is an agent wallet trading app where a Circle Agent Wallet buys paid market intelligence, attaches receipts and live source evidence to each decision, runs model review and risk checks, then executes approved actions on Hyperliquid testnet. If the primary review request fails, the backend uses a configured secondary model helper for the same review contract; the dashboard does not expose a separate provider panel.

Agent Diary gives AI agents a place to record what they actually did each day — without humans writing logs or managing API keys. Problem: Agents run hundreds of tool calls, but their work leaves no durable, readable history. Operators can’t see patterns (failing tools, workload spikes, incomplete tasks) at scale. Traditional APIs need signups, keys, and subscriptions — a poor fit for autonomous agents. Agentic AI: Agents submit normalized session reports (tools, tokens, timing, outcomes). The service aggregates signals and uses Claude Haiku to synthesize first-person diary prose grounded in that data. A separate reflect endpoint analyzes trends across days. Blockchain: All API access is gated by Circle Gateway x402 — agents pay USDC micropayments on Arc Testnet per request ($0.001 write, $0.0005 read, $0.01 reflect). No accounts or credit cards; payment is embedded in the HTTP flow (402 → pay → 200). This is developer API monetization for the agentic economy.

Open Agentic data feed accepting x402 payments, recording request/response traces to swarm feed. Simple demo gets prices - later implementation will perform deep analytics. Users/Agents (i.e. consumers) can ask for data that might be sourced from multiple feeds (handling their own provider subscriptions), but users only need to pay per call via x402.

SparkLead solves a real B2B sales gap: companies know their existing customers well, but have no systematic way to find which of those customers' competitors need the exact same product — most lead-gen tools optimize for volume over relevance, wasting most outbound spend. An autonomous agent (Claude Agent SDK) reasons about a budget and decides which customers to check, then pays a Circle Agent Wallet's USDC, via x402, to query a service that runs a live Tavily web search and Nebius LLM reasoning pass to identify competitors with evidence and a confidence score — not a generic database lookup. Every payment settles on Base mainnet through Coinbase's CDP facilitator, so the agent's spending is a verifiable on-chain transaction, not a black-box subscription. Results land as a weekly Slack digest, with zero crypto exposure for the end user.

Yes. AI and blockchain! Problem. Reaching your own AI agent should be as easy as texting a friend, but the channels fall short. Signal needs a phone number, Telegram is unencrypted and centralized. Agentic AI. Amtomat lives in the decentralized Status messenger: no phone number, end-to-end encrypted, as easy as any chat. And it does not just reply, it acts. Powered by Nebius with tool-calling, the agent reaches the Amt for you and books an appointment. Blockchain. Each appointment is written as a tamper-proof, content-addressed confirmation on Swarm, verifiable by its hash. The messaging channel itself runs on Waku, a decentralized peer-to-peer network.

Every SME in Europe has the same problem — they spend weeks manually researching markets, and days processing cross-border payments through expensive, slow, opaque systems. NeuroEconomy solves both. It's an autonomous AI economy where a Claude-powered orchestrator agent hires five specialist research agents, pays each one in real USDC via Circle Agent Wallets, and delivers a full intelligence brief — in under 30 seconds. Here's what makes it genuinely different. The Circle Agent Wallet isn't a demo feature — it's the financial backbone. Every agent gets a real USDC transfer with a real transaction hash. Unused budget is automatically refunded. The entire payment trail is recorded and transparent. Then we go further — into supply chain. Type a product, a quantity, a shipping route. NeuroEconomy fetches live packaging prices from the web, compares freight quotes, calculates landed cost including import duties, and autonomously settles payment with the supplier in USDC. If the balance is insufficient, it stops, shows you the exact shortfall, and generates a PDF invoice either way — so there's always an audit trail. This is what programmable money enables. Not just faster payments — fully autonomous, AI-driven financial decisions with complete transparency.

AI agents run from a single datacenter IP that the web blocks the moment they scrape at volume — 429s, geo-walls, bans — and the usual fix, a proxy network, gates access behind accounts, KYC, and monthly contracts a bot can't sign. Bytomic Proxy removes the human from that loop: it turns a proxy egress into a paid HTTP service metered by x402, the open "HTTP 402 Payment Required" payment protocol. An autonomous agent discovers the price on a free /catalog, hits the proxy, receives a 402 challenge, signs a USDC micropayment from its Circle Agent Wallet, and retries; Circle Gateway verifies and settles the payment gaslessly on-chain (Base), and only then does the proxy fetch the target through its own IP and stream the response back — with the settlement tx hash returned as a receipt. The wallet is the access credential and the budget: every call is logged to an auditable spend ledger, and the agent stops at its cap. Agentic AI is the buyer (an autonomous Claude Agent SDK loop that shops, pays, and budget-caps); blockchain is the rail (x402 + USDC + Circle Gateway settlement) that makes sub-cent, keyless, per-request commerce between an agent and a service actually work.

Procurement teams routinely need to vet suppliers before committing to bulk orders, but the data that actually matters — sanctions status, compliance risk — usually sits behind a paywall, and deciding whether a given supplier is worth paying to screen is itself a judgment call. This agent makes that judgment call autonomously: it researches a procurement question via Tavily, reads the results, and decides whether free information is sufficient or whether a paid compliance check is warranted — extracting the actual candidate company name to screen if so. When it decides to pay, it doesn't simulate a payment. The agent calls a paywalled endpoint, receives a real HTTP 402 challenge with the price and merchant address, and settles via Circle Developer-Controlled Wallets with a real USDC transfer on ETH-SEPOLIA testnet. The endpoint verifies the transaction actually confirmed on-chain — correct merchant address, correct amount, correct asset — before unlocking anything; a fabricated or replayed payment reference is rejected. Only once payment is verified does the agent receive a live OpenSanctions screening of the company, with the on-chain transaction hash returned as proof of payment. The final recommendation is grounded strictly in what the screening actually returned — the agent does not fabricate financial or credit data the API didn't provide. This combines agentic reasoning (the pay/don't-pay decision, company-name extraction, answer synthesis) with a real, verified blockchain settlement rail (x402-style 402-challenge flow over Circle USDC) — autonomous research, autonomous spend decision, autonomous payment, gated real data.

Research topics publically for free, pay in escrow to keep results private. Integrates Circle SDK for wallet gas sponsoring, escrow, and passkey. Uses tavily and tokenfactory for off-chain search, and inference respecively.

The first platform that gives AI agents secure, consent-controlled access to your entire digital life.

Circle + Sema Agent Marketplace solves the problem of trust, payment, and fulfillment between autonomous agents. Seller agents authenticate with Circle Agent Wallets and publish safe, user-owned listings such as data packs, research, proof services, warm intros, or link/file deliverables. Buyer agents search the marketplace, compare price, proof, risk, and reputation, then approve and pay through an x402-gated endpoint using USDC. Sema adds shared semantic context for provenance, buyer requirements, verification, and reviews. The blockchain/payment layer is used for wallet-based agent identity, USDC checkout, receipts, and seller payout tracking.

Giftr is a local gift-experience marketplace where anyone can browse and buy handcrafted goods, restaurant vouchers, spa treatments, and creative workshops — all priced and settled in USDC via Circle's infrastructure. No accounts, no credit cards, no payment forms. Just a wallet address and a payment that clears in under one second.