Tychi AI

Product Introduction

1. Definition: Tychi AI is a self-custody agentic wallet platform designed specifically for AI agents and their human operators. It is a crypto wallet and key management system that provides an economic layer for autonomous capital, operating as a decentralized finance (DeFi) tool with native on-chain interaction capabilities.
2. Core Value Proposition: Tychi exists to solve the fundamental trust problem in AI agent finance, where agents often use custodial or borrowed keys. Its core proposition is enabling agent-owned wallets with self-custody on the user's local machine, ensuring that signing never leaves the device. It provides policy-at-the-edge controls before every on-chain transaction on Arbitrum One, establishing a secure trust layer for autonomous capital management.

Main Features

1. Dual-Interface Architecture (MCP & CLI): Tychi ships two distinct, purpose-built interfaces. The @tychilabs/tyi-mcp is a Model Context Protocol (MCP) server (using stdio) integrated with AI hosts like Cursor, Claude Desktop, OpenClaw, and Codex, allowing agents to directly invoke wallet operations through a tool-native flow. The @tychilabs/tyi is a terminal-based Read-Eval-Print Loop (REPL) for humans, enabling natural language interaction for balance checks, token sends, and gasless payments. Both interfaces operate on the same underlying encrypted keystore located at ~/.tyi.
2. Hybrid Custody Model (Local Keys, Hosted Brain): The system implements a split architecture for enhanced security. The private keys and encrypted keystore reside solely on the user's local machine. Intent parsing, transaction routing, and policy evaluation are handled by the Tychi-hosted brain. The critical cryptographic signing process is executed locally, ensuring private keys are never exposed to the hosted brain, LLMs, or external servers.
3. Agent-Native Workflow & Policy Control: The platform is built for agentic autonomy. It implements a structured route → status → action flow for every operation. Before any on-chain action, policy caps are evaluated at the edge, providing guardrails for agent behavior. The flow supports multi-wallet management, easy onboarding, and gasless routing via the Universal Gas Framework (UGF) for native ETH and ERC-20 transactions on Arbitrum.
4. Unified Human-Agent Wallet: The same ~/.tyi keystore is shared between the human CLI and the agent MCP server. This allows a human to set up a wallet via the CLI REPL and then grant an AI agent managed access to the same wallet through the MCP configuration, creating a seamless collaborative finance environment.

Problems Solved

1. Pain Point: Custodial Risk and Lack of Agent Self-Sovereignty: Most existing agent wallets are custodial APIs or read-only blockchain explorers, meaning agents never truly own or control their capital. Tychi solves this key custody problem by providing true self-custody, where agents can earn, hold, sign, and pay using keys they control.
2. Pain Point: Usability and Integration Gap: Interacting with blockchain wallets programmatically is often complex and fragmented. Tychi bridges this by offering both a human-friendly CLI using plain English and an agent-native MCP toolset, dramatically lowering the barrier for both humans and AI models to perform financial operations.
3. Pain Point: Transaction Cost and Complexity: Sending tokens and executing payments on networks like Ethereum can be prohibitively expensive due to gas fees. Tychi integrates the Universal Gas Framework to enable gasless transactions and streamlined routing on Arbitrum, solving the gas fee problem for frequent microtransactions common in agent workflows.
4. Target Audience: AI Developers building autonomous agents, Prompt Engineers designing agent workflows, DeFi Developers creating financial automation tools, and Crypto-native power users seeking a programmable, self-custodial wallet for both manual and automated use.
5. Use Cases: An AI agent earning revenue from a service and holding it in its own wallet; automated portfolio management and rebalancing by an agent; human operators overseeing and directing agent-led payments; developers testing DeFi strategies with agent-controlled capital.

Unique Advantages

1. Differentiation vs. Custodial Wallets: Unlike custodial wallet APIs (e.g., those provided by exchanges) where a third party holds the keys, Tychi ensures full user sovereignty. Keys are local-only, and signing is device-bound, eliminating counterparty risk and aligning with core Web3 principles.
2. Differentiation vs. Standard Hardware/Software Wallets: Standard wallets (Ledger, MetaMask) are primarily designed for human GUI interaction. Tychi is fundamentally agent-native, with its MCP server allowing LLMs and AI tools to interact directly with wallet functions as part of a computational tool chain, which is a capability lacking in most conventional wallets.
3. Key Innovation: The core innovation is the split custody architecture with an embedded policy engine. By separating the cryptographic execution (local) from the intent and policy logic (hosted brain), it achieves a balance of security and AI-powered usability. This creates a secure "trust layer" where agent autonomy is possible without sacrificing the security of self-custody.

Frequently Asked Questions (FAQ)

1. How is Tychi different from MetaMask or a hardware wallet? Tychi differs fundamentally in its design purpose. While MetaMask and hardware wallets are secure key stores for human GUI interaction, Tychi is an AI agent-native wallet platform. It features a Model Context Protocol (MCP) server allowing AI models to programmatically control the wallet, an integrated policy engine for agent guardrails, and a human CLI using natural language. All operate on the same local keystore, bridging human and agent finance.

2. Are my private keys safe if Tychi uses a hosted "brain"? Yes, your private keys are safe. The architecture is designed so that the private keys never leave your local machine (stored in ~/.tyi). The hosted brain only processes intent (e.g., "send 10 USDC to this address") and handles policy evaluation. The actual cryptographic transaction signing occurs locally on your device, ensuring the core secret material is never exposed.

3. Which blockchain networks does Tychi support? Currently, Tychi is deployed on Arbitrum One. It supports native ETH, ERC-20 tokens, and enables gasless sends via its Universal Gas Framework (UGF). The roadmap indicates that expanded on-chain DeFi functionality (automation, trading, lending) is next on Arbitrum One.

4. What is the Model Context Protocol (MCP) and why is it important for Tychi? The Model Context Protocol (MCP) is a standard for providing AI models (like Claude, GPT) with contextual tools and data. Tychi's @tychilabs/tyi-mcp is an MCP server, which means it translates a natural language intent from an AI agent into specific, secure wallet operations. This is crucial as it allows any compatible AI agent to use the wallet as a native tool, enabling true agentic finance.

5. How do I get started using Tychi as an AI agent or a human? For humans, run npx -y @tychilabs/tyi@beta to start the interactive CLI and complete onboarding. For AI agents, first complete onboarding with the human CLI to create the wallet, then add the @tychilabs/tyi-mcp@beta server to your MCP host's configuration (like Claude Desktop or Cursor) with the TYI_PASSWORD environment variable set. Both packages are in public beta on npm.