jin-bo/agentao

Local-first, governed AI agent runtime for Python — embed it in your app, or run it as a CLI or ACP server. Permissions, MCP, memory and audit replay built in.

What it solves

Agentao provides a secure, governed runtime for AI agents. It addresses the risks of using autonomous agents by enforcing strict permission controls, ensuring transparency in decision-making, and providing a structured way to manage agent memory and tool access without relying on external cloud services.

How it works

It acts as a local-first harness for Python applications. When you embed it, you define the agent's environment, memory, and capabilities. It supports multiple permission modes (from read-only to full-access) and uses standard protocols like the Model Context Protocol (MCP) to connect agents to external tools. Every action taken by the agent is logged, and the system provides real-time visibility into the agent's "thinking" process.

Who it’s for

This project is designed for developers who want to build or integrate AI agents into their own software while maintaining control over security and data privacy. It is also suitable for power users who want a feature-rich, local-first CLI agent that supports persistent memory and custom skill sets.

Highlights

  • Governed Execution: Includes four permission modes and optional macOS sandboxing to prevent unauthorized actions.
  • Local-First Architecture: No external databases or cloud dependencies; everything runs within your local environment.
  • Full Observability: Real-time streaming of the agent's reasoning chain and comprehensive session logging.
  • Extensible: Supports custom skills, plugins, and integration with the Model Context Protocol (MCP).
  • Persistent Memory: SQLite-backed memory system that allows agents to recall information across different sessions.

Related

  • Project
    canyonroad/agentsh

    A secure, policy-enforced execution gateway for AI agents that intercepts and governs file, network, process, and database activity at runtime.

  • Project
    ZJU-REAL/HugAgentOS

    HugAgentOS is an open‑source, self‑hosted “AgentOS” that lets AI agents retrieve knowledge, run code, and complete real‑world tasks. It uses a domain ontology as a control plane for rule‑based, traceable execution, and offers chat, private RAG, sub‑agents, skill extensions, sandboxed code, multi‑layer memory, automation, and a web UI. Install via a one‑command script (SQLite) or Docker‑Compose (PostgreSQL/Redis, optional Milvus/Neo4j). Community Edition is Apache‑2.0; Enterprise adds multi‑tenant governance.

  • Project
    truffle-ai/dexto

    An open agent harness and orchestration layer that turns LLMs into stateful agents with persistent memory, tool integration via MCP, and a built-in production-ready coding agent.

  • Project
    EXXETA/exxperts

    A local-first AI agent framework that uses approval-gated memory to create persistent AI colleagues that learn and remember only what the user explicitly approves.