Skip to content

Overview

Assegai is a desktop application (Electron) that serves as a secure runtime for AI Agents that interact with blockchain networks. It solves the “unsafe agent” problem by sandboxing execution and enforcing a human-in-the-loop approval process for all financial transactions.

Unlike standard scripts that run directly on your host machine, Assegai agents run inside Docker containers.

  • Isolation: Agents cannot access your local file system (except their designated workspace).
  • Resource Limits: CPU and Memory usage are capped per agent configuration.
  • Network: Agents operate on a custom bridge network with restricted access.

Agents do not hold private keys. Instead, they use the @assegailabs/sdk to request signatures.

  1. Agent: Calls sdk.requestTransaction(...).
  2. Sandbox: Intercepts the request and pauses the agent.
  3. UI: Displays a transaction popup to the user with parsed data.
  4. User: Clicks “Approve” or “Reject”.
  5. Sandbox: If approved, signs the transaction via the connected wallet (e.g., MetaMask, Rainbow, or a local Anvil node) and returns the hash to the agent.

To prevent API key theft, Assegai includes a local proxy for LLM providers.

  • You configure your OpenAI or Anthropic API keys in the Assegai App Settings.
  • Agents use the SDK methods callClaude() or callOpenAI().
  • Assegai injects the key and forwards the request to the provider.
  • The Agent never sees the actual API key.

Assegai uses CAIP-2 chain identifiers for cross-chain compatibility.

  • EVM Chains: Full support for Ethereum (eip155:1), Polygon (eip155:137), Base (eip155:8453), Optimism (eip155:10), Arbitrum (eip155:42161), and more.
  • Testnets: Sepolia (eip155:11155111), Base Sepolia (eip155:84532), etc.
  • Localnet: Integrated support for Foundry/Anvil (eip155:31337) for zero-cost testing.