Echo Agent

Echo Agent is a self-hostable agent runtime for long-running AI assistants and applications. Cognitive memory lets the agent recall relevant conversation summaries, user preferences, and project facts across sessions, skill self-evolution distills reusable Skills from execution traces, and permission policies check every tool call.

The runtime runs in three forms: an interactive terminal agent, a Gateway service, and an in-process embedding inside a host application. The Gateway exposes a dashboard, an HTTP and WebSocket API, messaging channels, webhooks, and scheduled tasks, and the CLI or TUI can attach to a running Gateway to share its models, memory, tools, and policies. For example, the Gateway can tidy data on a schedule and send the result to Feishu or Telegram.

Memory is layered into the current task, session summaries, long-term facts, and an archive. Hybrid retrieval recalls entries, environment memories decay over time, and conflicting facts are detected and updated. Candidate Skills pass content validation and a baseline evaluation: low-risk candidates that improve on the baseline are promoted automatically, high-risk ones go to human review, and promoted Skills can be rolled back.

EchoAgent Desktop brings the runtime into a local workbench where the agent works with files, a browser, and the desktop, and shows its execution steps and changes. It bundles the Eclipse Theia IDE with a coding agent, browser and computer use with per-action confirmation for sensitive steps, and WeChat binding to continue a session remotely. State such as sessions, memory, and configuration stays in the deployment environment.