A real agent loop
Streaming, native tool calling, automatic context compaction when the window fills, and append-only JSONL history under ~/.rex/sessions/.
Most agents stop at writing code. REX runs the whole loop — understand the codebase, plan the change, make it, run the check that proves it holds — and hands you the controls at every step that matters.
Streaming, native tool calling, automatic context compaction when the window fills, and append-only JSONL history under ~/.rex/sessions/.
Writes are atomic and arrive as a diff you can edit before approving. Conflict detection catches files that changed underneath; every write drops a checkpoint.
Keys are encrypted with the platform credential store — DPAPI on Windows — rather than dropped in a dotfile. Environment variables still work as a fallback.
A real browser view composited into the window: navigate, read the accessibility tree, click by reference, type, evaluate and screenshot.
Native backends per OS — Win32/PowerShell, AppleScript, xdotool/ydotool/AT-SPI2. Never auto-allowed: every action asks, every time.
A REX.md at the project root is read into every session, so what you explain once outlives the restart and the model switch.
Reads run freely. Writes show a diff. Anything that touches your machine, your screen or the network asks first.
For authorised engagements only. These stay disabled on every account and every plan until they're switched on for that specific person — the agent enforces it itself, from a claim in the signed session token.
The React renderer keeps no secrets and executes nothing. It sends intents over IPC and renders what streams back — so a compromised page has nothing worth stealing.
NSIS and portable builds for Windows, a hand-assembled .deb that builds even from a Windows host, a distro-agnostic tarball, and macOS targets wired up.