Note
Early, and used daily by several senior engineers. Two update channels,
switched in Settings → Advanced: Stable (default) gets a release once
alpha users have run it; Alpha gets every passing change on main, often
several a day. Leaving alpha never downgrades. Point factorai at work your
version control can recover.
You stopped writing most of the code. Your editor never noticed.
factorai is built the other way round: the unit of work is a session, not a file. You supervise, decide, review, and set the rules. Agents do the rest. Free, open source, community-driven: no company, no paid tier.
Each session is the real claude or codex CLI in a real terminal, and keeps
running when you look away. A dot says whether it is working, waiting for
you or stopped.
A routine is a saved prompt on a cron, per project. It runs in the background without taking a tab, and catches up on runs missed while factorai was closed.
Drag projects into groups you name, or sort by name or recent. Keyboard too.
Full-text search across every transcript, read in place from ~/.claude and
~/.codex. Nothing imported, nothing copied.
Changes, diffs and the commit graph sit beside the terminal, and keep up with an agent mid-edit.
No telemetry, no account, no server. factorai drives the CLIs you already logged into and never touches your credentials. Closing the window kills every agent it started.
Download from factorai.build or
Releases: .dmg for macOS,
.AppImage for Linux (glibc 2.39+), factorai-setup.exe for Windows, which
installs the Linux build into WSL 2. It updates itself and asks before
restarting. You need Claude Code or
Codex, already logged in.
Something look broken? See installation and troubleshooting. Building from source is in CONTRIBUTING.md.