Inspiration
Almost everything we search, ask and create online goes through a few centralized platforms. Every query gets logged, tied to a profile, and can be monetized or used to train proprietary models without our consent. That creates data-breach risk, hidden ranking decisions, and a chilling effect on what people feel free to look up. We wanted to see if search could work without a company in the middle, so we built DAISY, a Decentralized Artificial Intelligence Search for You.
What it does
DAISY is a trust-minimized search marketplace on Algorand. A user posts a query and escrows a fee in the DAISY token, an Algorand Standard Asset. An independent AI/indexer node sees the query, generates an answer, and submits it on-chain with a content hash, a URI and a signed receipt. When the user accepts the response, the smart contract pays the node automatically. The receipt lets anyone verify who answered and that the answer wasn’t altered. Payloads stay off-chain and only their hashes go on-chain. Anyone can run a node and earn DAISY.
How we built it Smart contract: an ARC-4 app written for Algorand that handles escrow, query records and settlement. Token: a custom DAISY ASA with 0 decimals for fees and rewards. Deployment: a script built on AlgoKit Utils that creates the token, deploys the app, funds it, and opts it into the ASA through an inner transaction. AI node: a Python worker that polls the contract for new queries, answers them with Google Gemini, and submits the response. Client library: generated helpers for algod and app-call composition, plus a prompt script for posting queries. Dev tooling: top-up scripts that refill ALGO and DAISY on the test accounts. We tested on Algorand LocalNet. Challenges we ran into Getting the economics right on Algorand. Inner transactions need their fees covered by the outer transaction, and the app account needs extra ALGO for its minimum balance before it can opt into the ASA. Keeping answers short enough to fit within ABI argument limits when they’re submitted on-chain. Handling version differences in AlgoKit and the Algorand SDK, which changed how clients and accounts are created. Designing around on-chain storage costs, which pushed us toward keeping data off-chain and storing only hashes. Accomplishments that we’re proud of A working end-to-end flow on LocalNet, from posting a paid query through to an AI answer, a receipt and a payout. Cryptographic receipts that make answers verifiable instead of something users have to trust. A clear separation of contract, node and client, so others can swap in their own indexer or frontend. What we learned How to build and deploy ARC-4 contracts with AlgoKit and design a token-based escrow on Algorand. How inner transaction fees, resource population and minimum balance requirements shape contract design. That decentralizing a service is as much about incentives and verifiability as about the technology. What’s next for DAISY_Chain Implement the timeout and reclaim path so users can recover escrow if no acceptable answer arrives. Test multiple competing AI nodes on TestNet, using VRF and PPoS for fair selection and validation. Store query and prompt logs on decentralized storage like IPFS or Nillion and keep only CIDs on-chain. That removes the need for BoxMaps, keeps the minimum balance low, and makes the system scale. Add a relayer so users can pay entirely in DAISY while the relayer covers ALGO fees. Build a web frontend for posting queries and reviewing receipts.
