Krait: Verifiable AI Trading Agent 0G Hall Proposal
Project Overview and Objectives
Krait is an autonomous artificial intelligence trading agent explicitly designed to think and trade like a human professional, continuously operating across Web3 crypto and traditional finance forex, commodities, and indices. Currently, AI decentralized finance trading suffers from the black box problem: users hand over capital to algorithms without knowing exactly why trades are executed or if the code is genuinely making the decisions.
Our objective is to build the first mathematically and cryptographically verifiable fund manager that replicates the deep logic and context awareness of a human professional. Human traders suffer from severe limitations: emotional tilt, fatigue, slow reaction times, an inability to monitor multiple markets simultaneously, and irrational deviation from their own strategies. Krait solves this by thinking like a human but executing with machine perfection, completely avoiding these fatal flaws. By executing through non custodial smart contracts, using machine learning to detect market setups, and leveraging the 0G Network for verifiable LLM reasoning, Krait ensures every trade is emotionless, instantaneous, authorized, explainable, and permanently auditable.
Expanded Capabilities
Krait treats blockchain protocols and traditional Web2 brokerages as a single unified liquidity pool. Its core capabilities include:
-
Omnichain and Multiple Market Execution: Trades natively on Sui, Solana, Base, and Monad for crypto spot, perpetuals, meme coins, and yield farming, while concurrently integrating with brokerages like OANDA for forex majors, gold, silver, and indices.
-
10 Execution Modes and 27 Strategies: Capable of everything from smart dollar cost averaging and arbitrage to high frequency meme coin momentum catching and macro news trading.
-
Continuous Vision Quality Assurance Loop: The agent physically renders charts of its own technical analysis drawing order blocks and gaps, takes screenshots, and uses a Vision LLM to visually verify its own math before executing a trade.
-
Omni Channel Interfaces: Controlled via a developer grade CLI, a unified Web dashboard, a custom TradingView extension, and a natural language chat system across Telegram, WhatsApp, and Web.
-
Smart Capital Transfer: Features a Simple Mode where users deposit USDC into their preferred home chain. Krait securely uses Circle Cross Chain Transfer Protocol versions 1 and 2 to natively burn and mint USDC across chains, ensuring absolute capital efficiency with Wormhole acting strictly as a fallback. The user only ever deposits USDC once, and the AI handles all movement natively through sponsored transactions to automatically chase the best yield and arbitrage opportunities.
Technical Architecture
Krait features a Three Layer Intelligence Engine built over a robust execution layer:
-
Technical Pattern Engine: A deterministic mathematical engine running Smart Money Concepts to parse complex market structures.
-
Machine Learning Scorer: An XGBoost and ONNX layer that predicts the probability of technical setups based on walk forward validated historical data.
-
Dual LLM Reasoning: A fast lightweight Qwen model from 0glabs runs constantly in the background as a Sentinel scanning breaking news and macro data, while a heavy reasoning model GLM 5 is only woken up to make the final verifiable trade decision.
-
Non Custodial Vaults: Execution happens through specialized smart contracts on four chains. The agent cannot withdraw funds; it operates within strict hardcoded risk bounds like max 3 percent daily loss limits.
Integration with 0G Infrastructure
Krait maximizes integration depth by utilizing four distinct 0G components to create a completely verifiable proof chain:
-
0G Compute: Every final trade decision runs through GLM 5 via 0G Compute. This generates a cryptographic proof known as an X Inference Proof, proving mathematically that the LLM generated the specific trade signal.
-
0G Storage: Every trade decision, along with its specific reasoning context and inference proof, is persisted to decentralized 0G storage. This creates a permanent auditable history of the agent’s logic.
-
0G Chain: Krait deploys a registry contract KraitRegistry.sol on the 0G Chain. Every trade logs a hash anchored on chain, finalizing the verifiable proof chain.
-
Agent ID: An ERC7857 Agent NFT proves ownership of the Krait agent and stores encrypted strategy configuration metadata.
Flow: AI Inference Compute to Logic Recorded Storage to Hash Anchored Chain.
Team Background and Experience
Krait is developed by a highly capable solo developer with extensive experience shipping production quality code. The project codebase features over 500 tests across 10 packages, bridging three distinct smart contract languages Move, Solidity, and Anchor with complex off chain TypeScript and Python machine learning pipelines.
Funding Requirements and Milestones
Funding Request: $50,000 for infrastructure, security auditing, multiple chain operations, and mainnet scaling.
Milestones:
-
Milestone 1 Month 1: Complete testnet paper trading loop. Migration to official 0G SDKs for Compute and Storage. Deployment of KraitRegistry.sol to 0G testnet.
-
Milestone 2 Month 2: UI completion. Release web dashboard showing agent state and trade history linked to 0G explorer proofs. Finalize TradingView integration for visual verification.
-
Milestone 3 Month 3: Mainnet deployment. Live execution across Base, Sui, Solana, and Monad. ERC7857 Agent NFT minting framework fully integrated.