Insight-AI
Autonomous Telegram bot predicting whale moves via x402, Chainlink, Uniswap v4, ENS & LI.FI
Demo
About
Insight AI is an autonomous crypto intelligence agent delivered through Telegram. When a whale moves large amounts of ETH, the bot: - Detects the transaction and resolves the wallet's ENS identity (name, avatar, social profiles) - Queries Chainlink oracles on-chain for verified, tamper-proof ETH/USD pricing - Analyzes Uniswap v4 pools for whale swap patterns and Hook behavior (TWAMM, LimitOrder, Dynamic Fee) - Scans cross-chain activity across 6 EVM chains via LI.FI bridge intelligence - Uses Google Gemini AI to autonomously decide which premium APIs to purchase - Pays for data via x402 micropayments (USDC on Ethereum) — no subscriptions, pay-per-insight - Synthesizes everything into a single actionable report delivered to Telegram The AI agent operates with full autonomy — it evaluates oracle confidence, calculates cost-benefit ratios, and only spends when the data is worth purchasing.
How it's made
Insight AI is a Bun/TypeScript monorepo running a Telegraf Telegram bot in front of an Express server that exposes x402-gated endpoints; when a whale-monitoring service flags a large ETH transfer, a Google Gemini agent weighs the transaction's size and volatility and decides which premium data sources are worth paying for in USDC via x402 — no subscriptions, no API keys, just per-call micropayments. It then pulls Chainlink oracle prices (with on-chain confidence/staleness checks, cached and snapshotted via Prisma/Postgres), resolves the whale's identity through ENS (`lookupAddress`/`getText`, backed by a curated whale mapping for addresses without public ENS records), checks Uniswap v4 pools for Hook activity (TWAMM, LimitOrder, Dynamic Fee) to catch swap patterns, and layers in LI.FI cross-chain data across six EVM chains to read bridge behavior — synthesizing all of it into one Telegram report. The hackiest bit is letting the LLM make real (if tiny) autonomous spending decisions, which meant building confidence-threshold and cost-ceiling guardrails so Gemini couldn't just buy every data source for every alert.