Skip to main content

​
The problem today

AI agents that want to do things on-chain face a hard choice:
  1. Hold private keys - Dangerous. If the agent is compromised, funds are gone.
  2. Use centralized services - Defeats the purpose. You’re trusting a third party.
  3. Require human approval for everything - Slow. Kills automation.

​
How Umi solves it

Umi’s architecture is perfect for AI agents because of two things: Zero-trust signing The agent never has complete access to the keys. Signing requires both the agent’s share and the Ika network’s share. Even if the agent is compromised, an attacker can’t drain funds without also compromising the network. Programmable guards You can set rules on what the wallet can do:
  • Spending limits per transaction or per day
  • Allowlists of addresses that can receive funds
  • Blocklists of contracts or addresses
  • Rate limiting to prevent rapid draining
  • Multi-sig requirements for large transactions

​
The Umi MPC Server

Umi provides a dedicated MPC server that agents can connect to for signing operations. This is the infrastructure layer that makes autonomous cross-chain agents possible. What the MPC server provides:
  • Secure signing endpoint for AI agents
  • Policy enforcement before any signature is produced
  • Multi-chain support (Sui, Ethereum, Bitcoin, Solana, etc.)
  • Audit logging of all agent actions
How agents connect: Agents authenticate with the Umi API and receive scoped access to sign transactions on behalf of a Umi account. The API handles:
  • Session management
  • Transaction building
  • Signature coordination with the Ika network
  • Broadcasting to destination chains
The agent never touches raw keys. It sends transaction requests, the MPC server enforces policies, coordinates the zero-trust signing, and returns the result.

​
API for agents

The Umi Agent API lets any program sign transactions using a Umi account. Authentication: Agents authenticate using API keys scoped to specific accounts and policies. The account owner controls what the agent can do. Signing flow:
  1. Agent builds a transaction (destination, amount, chain)
  2. Agent sends signing request to Umi API
  3. Umi validates against the account’s policies on chain
  4. If approved, MPC signing produces the signature
  5. Transaction is broadcast and result returned
Example operations:
  • Sign and send ETH transfers
  • Sign and send BTC transactions
  • Sign and send SOL transactions
  • Sign arbitrary messages for authentication
  • Query account balances and history

​
Use cases

Trading bots An AI agent that trades across chains based on market conditions. Set a daily spending limit and let it operate autonomously within those bounds. Portfolio rebalancing Automated rebalancing across your multi-chain portfolio. The agent can move assets between chains but only to addresses you’ve approved. DeFi automation Yield farming, liquidity provision, and other DeFi strategies that span multiple chains. Guards ensure the agent stays within defined parameters. Payment processing Automated payment systems that can send across any chain. Rate limits and allowlists keep things safe even if the system is compromised. DAO operations Programmatic execution of DAO decisions across chains. Multi-sig requirements ensure critical operations need multiple approvals. AI assistants Give your AI assistant the ability to execute transactions on your behalf. “Send 0.1 ETH to alice.eth” becomes possible with proper guardrails.

​
Setting up an agent

  1. Create a Umi account (or use your existing one)
  2. Configure signing policies (limits, allowlists, etc.)
  3. Generate API credentials for your agent
  4. Connect your agent to the Umi MPC server
  5. The agent can now operate autonomously within the defined boundaries
The agent signs transactions through Umi’s MPC infrastructure. It never has raw private key access. If something goes wrong, the damage is limited by your policies.

​
The bigger picture

This is what makes Sui + Ika special for the AI era:
  • Sui coordinates everything
  • Ika provides the zero-trust custody
  • Umi’s MPC server provides the agent interface
  • Programmable policies define the boundaries
  • AI agents operate freely within those boundaries
Safe, autonomous, cross-chain AI agents. Not possible before. Possible with Umi.