tuchg/Lucarne

What it solves

Lucarne eliminates the need to constantly monitor local AI agents. It allows users to step away from their computers while receiving real-time notifications, providing approvals, and answering clarifying questions via mobile messaging apps, ensuring that long-running agent tasks are not stalled by the need for human intervention.

How it works

Lucarne acts as a lightweight resident background process (daemon) that connects local AI agent CLI processes (such as Claude, Codex, Gemini, and others) to mobile communication channels. It uses channel adapters for Telegram and WeChat to push progress updates to the user and route replies back to the agent.

Key mechanisms include:

• Quote-to-Route: In WeChat, users can quote a notification and reply to it, which automatically restores the matching agent session context.
• Mobile Console: In Telegram, users can use a panel to create workspaces, bind agents, and manage sessions using Forum Topics.
• Zero-Intrusion Setup: It monitors existing agent processes without requiring hooks, skills, or project changes.

Who it’s for

Developers and power users who run autonomous or semi-autonomous AI agents locally on their machines but want to manage and monitor them remotely via their phones.

Highlights

• Multi-Channel Support: Integrates with Telegram and WeChat for notifications and control.
• Broad Agent Compatibility: Supports a variety of agents including Claude, Codex, Gemini, Copilot, Pi, and Grok Build.
• Actionable Notifications: Turns permission approvals and clarifying questions into mobile events that can be approved or denied with a tap.
• Low Overhead: Designed as a high-performance, low-memory resident process that automatically releases idle agents.
• Session Management: Supports forking sessions, interrupting work, and viewing historical session history.