From b17a8b3c09a88929d6ee618f80d654013975675c Mon Sep 17 00:00:00 2001
From: Alfonso Sastre The web UI, recorded live: a real local (Ollama) session implementing A real, unedited run — the request is plain English, not Lex. No signature, no syntax, no prior Lex knowledge. Those tools end when the model
+ says it's done and the diff looks right to you. lex-code ends when An editor assistant runs
+ with whatever access your editor has. lex-code's capability grant is enforced by the Lex
+ VM itself — an effect it wasn't granted, it cannot perform, full stop, not "was told not
+ to." It isn't trying to replace your daily
+ driver. It's a plain CLI any agent — Cursor, Claude Code, Codex, a CI step — can shell out
+ to specifically for Lex, the one thing it holds to a stricter bar than "the diff looks
+ plausible."
-
- zip, then its full event Trail.Why not just use Cursor, Copilot, or Claude Code directly?
+ Different confidence signal
lex check
+ passes and a declared acceptance does too — a result you can act on without reading the
+ diff, not a transcript claiming success.An enforced sandbox, not a polite one
Not a competing IDE
Try it
--verify, --bar, --multi), the web UI, and the
typed-issue workflow in depth: the
README.
+
+ Prefer a UI: the same session, in the web frontend, with a live + Trail tab.
@@ -184,6 +206,10 @@