Loops HouseOralys Signal Market

NEAR HACKATHON
Links
Team
1 member- JOOwner
John Goldschmidt
Overview
A lot of valuable information is shared only in person, and today it gets passed on almost anecdotally. Yet hundreds of people in the room have phones that can corroborate what's said and stream it in real time to listeners: agents looking for alpha or gathering signal. Those agents listen for what they need and pay the people recording and corroborating it.
On Oralys, people at events record the latest on technology, policy and markets. With Oralys Signal Market, a buyer agent places a standing keyword request, such as "NVIDIA" OR ("Clarity" AND "Act"), and sets its own lookback window (the last hour by default) and a fixed price per excerpt (0.01 NEAR in the demo). When a recording in that window matches, the agent pays in real time for that excerpt, the matching sentence plus one on each side, through an escrow contract on NEAR, Oralys's home chain.
How NEAR fits: the buyer's deposit caps total spend. Matching runs off chain and is attested by a seller oracle that signs the keyed transcript hash and quality score. The contract verifies that ed25519 signature on chain, enforces the buyer's minimum quality score, pays each excerpt once per request, and lets anyone refund the unspent deposit after the deadline. Transcripts never go on chain.
On testnet, 15 transactions show 6 paid excerpts, a 0.04 NEAR refund, and rejects for a forged signature, an out of window match, a low score and a double payment.
In the one verified run on NEAR AI Cloud, glm-5.3-flash answered stage 02 of the v1 buying agent, which picks what to buy, inside its TEE, and the reply's provider_tee signature checked out, matching the rules engine. Keyword matching makes no TEE claim. Limits: the reply body is not saved, and report and reply are not proven to come from the same serving instance.
Recorder tokens are not built. All data is SAMPLE DATA. Oralys is an existing product; only the NEAR piece was built in the hackathon window, by AI coding agents reviewed by the team.