Agents Overview

Agents are containerized programs that run in Docker with controlled access to blockchain networks and AI APIs. Each agent runs in strict isolation with defined permissions, resource limits, and a mandatory “Human-in-the-Loop” approval flow for transactions.

Agent Capabilities

Agents are designed to be autonomous but safe. They can:

• Query Blockchain Data: Read data from whitelisted RPC endpoints (e.g., getting balances, simulating transactions).
• Request Transactions: Propose transactions that require user approval to execute.
• Use AI Models: Access configured LLMs (OpenAI, Anthropic) via the secure proxy.
• Persist Data: Read and write to a dedicated /agent-workspace volume.
• Log Activity: Send structured logs to the user dashboard.

Agent Structure

A minimal agent requires the following file structure:

my-agent/
├── assegai.json       # Manifest configuration
├── index.js           # Main entry point
├── Dockerfile         # (Optional) Custom build steps
└── package.json       # Node.js dependencies