1jehuang/jcode v0.52.0 on GitHub

ChatGPT web models, faster /model search, and a calmer TUI

New browser-backed ChatGPT web route adds gpt-5.6-pro[web], letting Pro-tier web models run through your logged-in browser
/model fuzzy search is dramatically faster and no longer lags while typing
Opening the / slash-command palette no longer shifts the screen, and the menu renders as an overlay

Platform-API-only GPT Pro models now appear in the /model picker
All providers now expose their complete reasoning effort ladders
macOS menu bar sessions are labeled by their active work and show only your root sessions
New display.overscroll_status config (off/on/overscroll) and overscroll now requires a gesture starting at the transcript bottom
New JCODE_PERF_TIER environment variable forces a performance tier for one invocation
Auto-poke stops after consecutive provider guardrail refusals instead of retrying forever
Tool calls show a short intent line by default, with technical detail behind display.tool_call_details
Long-idle sessions periodically release retained memory back to the OS

Copy uses native Linux and macOS clipboards before falling back to OSC 52
Update checks authenticate to GitHub when possible, avoiding rate-limit failures on shared IPs
/fast set to off now persists instead of reverting to the default
Memory tool results are scoped to the session working directory instead of leaking across projects
PascalCase OAuth tool aliases now resolve inside batch subcalls
Windows server spawn no longer times out prematurely on slow machines
Desktop attach no longer blocks on a busy session
PowerShell install scripts no longer break from a UTF-8 BOM
Logging in no longer activates the same profile twice
Split-pane launches keep their elapsed-time display
Bash commands no longer spill onto a second line when an intent is shown

Full changelog: v0.51.1...v0.52.0

Platform availability