argeneau12e/kairos-tx

KAIROS is a TypeScript project that solves a specific problem in Solana blockchain development: how to submit transactions in a way that is reliable, cost-effective, and aware of current network conditions. Most tools treat sending a transaction as a single fire-and-forget action. KAIROS treats it as a five-stage pipeline, each stage backed by live data and an AI reasoning layer. The core of the system is a bundle submission pipeline that uses Jito, a service that lets you submit groups of transactions together for priority processing. Instead of using a fixed tip amount, KAIROS fetches live tip floor prices from the Jito API and lets an AI agent running Llama 3.3 70B via Groq decide how much to tip based on current network health. If network conditions are poor and the estimated chance of a transaction landing falls below 30 percent, the system holds off on submitting and waits for conditions to improve. Network health is tracked using four live signals pulled from a Yellowstone gRPC stream: the gap between when a transaction is processed and when it is confirmed is one of them. This stream delivers real-time blockchain events rather than requiring repeated polling. The system also runs a pre-flight simulation before spending any fees, which catches failures before submission. Every bundle submission is logged with a tip efficiency score that compares what was paid against the 75th percentile tip at the time. After each run, an AI-generated prose report summarizes what the system observed. A live terminal dashboard shows network health sparklines, a Jito leader schedule, a blockhash countdown, and the bundle pipeline in real time. The project was built for a Superteam Nigeria infrastructure challenge and requires API keys for Groq and a Yellowstone gRPC node.