Echo Agent
A self-hostable, long-running AI agent for personal or team automation, with four-layer cognitive memory, a forgetting curve and contradiction detection, plus LLM-based approval for high-risk actions such as running commands or files.
Echo Agent is a self-hostable AI agent designed to run continuously rather than answer one-off questions. It carries memory across sessions using a four-layer cognitive memory structure with automatic decay and contradiction detection, aiming to avoid unbounded memory growth over long-running use, and it can generate candidate improvements to its own skills from real execution traces, which take effect only after passing evaluation and can be rolled back.
It unifies access across 14 channels, a CLI, a gateway API, webhooks, cron, and messaging platforms including Telegram, Discord, Slack, WeChat, WeCom, Feishu, DingTalk, QQ, WhatsApp, email and Matrix, sharing one state across all of them. High-risk tool calls such as command execution or file operations go through unified LLM-based approval and explanation, credentials are stored encrypted, and execution is logged for audit.
It natively supports MCP, the A2A protocol, multi-model routing and task scheduling, installs with pip install echo-agent[all], and has a separate Tauri 2 desktop client with a React and TypeScript interface over a Rust runtime.