fuyuxiang/echo-agent

面向长期运行与桌面协作的开源 AI Agent，提供常驻 Runtime、桌面 Agent、长期记忆与持续学习能力，可用于个人助理、自动化任务与本地工作流。

Echo Agent – Quick Overview

What it is – Echo Agent is an open‑source agent runtime that lets you run a continuously‑operating AI assistant on your own hardware. It adds three key AI‑focused capabilities:

Long‑term memory – the agent can recall past conversation summaries, user preferences and project facts across sessions.
Self‑evolving skills – after completing a task the runtime analyses the execution trace, extracts reusable “Skills”, evaluates them and can automatically promote low‑risk skills for future use.
Pluggable execution – tools, external MCP services, plugins and custom permissions can be hooked into the runtime, and the whole system can be exposed via a Gateway service or embedded in another application.

How you run it

Mode	Command	Typical use
Local (terminal)	`echo-agent run`	Quick interactive debugging or single‑shot tasks
Gateway service	`echo-agent gateway`	Long‑running daemon, dashboard, HTTP/WebSocket API, scheduled jobs
Embedded	EchoAgent Desktop (built with Tauri)	Integrated into a desktop app, giving the agent access to files, browsers, IDE, etc.

All three modes share the same core runtime; the Gateway simply wraps it with a web UI and API.

Core building blocks

Component	Role
Agent Loop	Orchestrates model inference, tool calls and feedback until a task is finished
Session & Context	Holds per‑task state and assembles the context for each model call
Model routing	Connects to one or more LLM providers, with fallback logic
Tools / MCP	Executes local commands or forwards to external tool‑servers
Skills & Plugins	Reusable workflows that the agent can invoke automatically
Permission policy	Checks whether a tool call is allowed and logs the decision
Automation & Channels	Cron‑style jobs, webhooks, message‑queue entry points

Memory & Evolution

Cognitive memory – layered store (session, short‑term, long‑term, archive) with mixed‑retrieval and time‑based decay. Conflicting facts are detected and updated.
Skill self‑evolution – execution traces are mined for candidate skills; low‑risk candidates are auto‑promoted, higher‑risk ones go to manual review, and promoted skills can be rolled back.

Desktop client (EchoAgent Desktop)

Built with Tauri, Node.js, pnpm and Rust.
Provides an IDE‑like workspace (Theia), browser sandbox, and “Computer Use” that can manipulate the desktop after user confirmation.
Can bind to WeChat for remote continuation of a session.

Extensibility points

Extension	What you can add
Tools	New native commands or scripts
MCP	Connect to external tool‑service platforms
A2A	Inbound task ingestion (outbound delegation not yet supported)
Skills / Plugins	Custom reusable workflows
Models	Plug in any LLM service via the model connector
Channels	New message sources (e.g., Slack, Telegram)

Getting started (quick)

pip install "echo-agent[all]"          # install runtime + optional deps
echo-agent setup                         # run wizard to add an LLM API key
echo-agent run                           # start interactive REPL

For a persistent service use echo-agent gateway (or gateway install to register a systemd/user service).

Documentation & community

Full docs: https://fuyuxiang.github.io/echo-agent/
Dashboard UI at http://127.0.0.1:58123/ when the Gateway is running.
Discussions, issues and a Chinese QQ group for support.

License

MIT‑licensed core; desktop client includes third‑party components with their own licences.

Bottom line – Echo Agent is a self‑hosted, extensible AI‑agent platform that adds memory, skill‑learning and tooling to any LLM you connect, usable from the command line, as a service, or embedded in a desktop UI.