# Agentica

A CLI, local web and desktop AI agent product that runs entirely on the user's own machine, letting one person operate a team of specialist agent sessions with self-evolving skills.

Agentica presents the same product as a CLI, a local web UI and a desktop app, all backed by one local installation. A single terminal session is one agent; within a session, a lightweight "task" spins up a temporary subagent, and a heavier "delegate" starts a whole separate agent process for an independent job, with cross-terminal "peer" messages letting two sessions talk to each other, all without extra deployment steps.

Completed work is automatically compiled into reusable SKILL.md files, so the agent reads its own prior conclusions on similar tasks instead of starting over. Context compression writes oversized tool output to disk immediately, keeping only a short preview plus a file path in context, and evicts only the oldest entries once the window nears capacity, so no part of a long session is lost to a lossy LLM-generated summary. It also connects to WeChat, WeCom, Feishu and Telegram so a networked gateway agent can address any running session by name.

Website: https://shibing624.github.io/agentica/