mercury-agent

Why we included this project

Mercury is an agent you run on your own hardware and leave running continuously from a terminal or Telegram. The safety model is the part that stands out: a blocklist keeps dangerous commands like sudo or recursive deletes from ever executing, file access is confined to folders you allow, and by default the agent pauses to ask before each action unless you switch a session to Allow All. It also keeps a searchable SQLite-backed memory of earlier decisions and your preferences, so you can point it at a task and come back later without re-explaining context. That combination of always-on operation and explicit guardrails makes it a practical pick for developers who want an unattended assistant they can trust.