Hi Product Hunt 👋 Michele here again, and this is Inner Droid. We built it because we wanted AI agents that actually do things (answer email, move meetings, keep notes current) without handing them the keys to everything. An agent that can send an email can also send the wrong one. So the whole app is built around a gate: 🟡 Agents run as a team on your machine, on a heartbeat. They keep working while you're away. 🟡 Every sensitive action (an outgoing message, a file change, a shell command) stops at the gate with a #ref in an append-only ledger. You approve it or you don't. 🟡 You can grant an agent autonomy for one kind of action and revoke it later. 🟡 Memory, email, calendar and reminders, a meeting recorder with local transcription, and WhatsApp, Telegram, Discord and Slack are built in. 🟡 The database is encrypted on disk (SQLCipher). You bring the LLM: Claude, ChatGPT, OpenRouter or the OpenAI API. Only what a tool explicitly sends leaves your machine. It runs on macOS, Windows and Linux, and it's free for the whole beta. No card, no trial clock. Two things we'd love to hear from you: 1. Where would you draw the line between "just do it" and "ask me first"? 2. What's the first job you'd hand to an agent you could trust? We're here all day and we'll answer every comment.

Inner Droid

AI agents on your machine that ask before they act

AI agents on your machine that ask before they act