Tome – a security first desktop cockpit for coding agents
Hello everyone, long time lurker here. Firstly I would like to thank you for taking the time to read my post and (hopefully :) ) taking a look at my project.
Tome is a desktop app (macOS + Linux) that puts your coding agents (claude, opencode, pi), terminals, editors, and an AI assistant into one tiling workspace with a sandbox around the agents.
a short video showing how you can make loops or graph engineer using the flow tool
I use Tome everyday to orchestrate agents, build flows and generally learn new technology whilst building projects. One of my favorite features, "verbose mode", teaches the user how what they've built actually works before pushing to remote repos.
The part I think is actually different: agents run inside a containment cell, an OS sandbox (sandbox-exec on macOS, bubblewrap on Linux) whose only route to the network is an allowlisted loopback proxy. It's a boundary you can see and unlock on purpose. Every unlock and blocked host lands in a security event log.
New in v0.4.0: voice. Fully on-device transcription (Apple Speech on macOS, whisper.cpp as the offline fallback) — streaming, hands-free, and audio never leaves the machine. Talk to the assistant, talk over it to interrupt, and it answers back.
Also in there:
- An assistant that can list/read panes and type into terminals (auto-run is off by default)
- Flows: DAGs of agent nodes
- Mentor mode, an in-app git UI, a note vault ("brain"), workspaces
- MIT, macOS + Linux (the DMG is unsigned for now)
GitHub: https://github.com/zwaneldmz/tome
Release: https://github.com/zwaneldmz/tome/releases/tag/v0.4.0
I'd especially value feedback on the security model and the voice UX, those are the two things I think were hard to get right.
Once again thank you for your time!