Loops Housenearagent

NEAR HACKATHON
Links
Team
1 member- MEOwner
Meliksah Barsbay
Overview
Swap interfaces are visual: charts, popups, and icon-only buttons. A blind user on a screen reader often cannot tell what a swap will cost or what they will receive.
nearagent is an accessibility-first crypto assistant for NEAR Intents. It has one text box. You type or say a sentence in Turkish or English, such as "swap 2 NEAR to USDC", and it answers in short sentences that a screen reader reads aloud. Every quote is read in one fixed order: what you give, what you get, the guaranteed minimum, the fee, the time, and how long the quote is valid. Every step works from the keyboard. A confirmation is held on the server, works once, and expires with its quote. Amounts above 100 dollars are spelled out and confirmed first.
How NEAR fits in: NEAR Intents 1Click supplies token data, prices, and quotes in dry mode, so one sentence can price a swap across chains with no bridge screens. The NEAR Intents Agent API is built in for a policy-bound wallet: live swaps, withdrawals only to the owner's address, a daily budget, freezing in one step, and unfreezing only with a six digit code. A separate script sends NEAR from the owner account. NEAR AI Cloud is the first provider for parsing, for "did you mean" suggestions when speech garbles a token name, and for speech to text with Whisper in its TEE (experimental). OpenAI is an automatic backup.
Current status: dry mode is the default and works with the real 1Click API. On NEAR mainnet, through the Agent API's demo server, an agent was created, funded with 5 USDC, frozen and unfrozen. On 5 November it made one real swap, 4.9 USDC to 1.207 wNEAR, confirmed from the page with the NVDA screen reader. Proof links are in the README. On 9 November it withdrew 0.1 NEAR to the owner account. The NEAR send script was only tested locally. NEAR AI Cloud parsed 43 of 44 test sentences itself. One timed out, and the OpenAI backup got it right. That set also shaped the prompt. It also fixed 5 of 5 garbled sentences, with a 1.5 second median.