v0.2.4

v0.2.4

New

• /context-usage command (CLI) — progress bar with percentage and used/total tokens
  for the active model's context window, color-coded amber above 60% and red above 80%.
• Context ring (VS Code) — a circular gauge beside the composer toolbar shows how full
  the context window is, colored by pressure. Hover for the numbers; click for a Session
  Info popup with a used/total breakdown and progress bar. Shown for models that report a
  context window.
• Collapsible providers in the model picker — fold provider groups in both apps.
  CLI: focus a heading and use Enter/←→ (Shift+←→ folds/unfolds all). VS Code: click a
  header, or use the Collapse all / Expand all toggle in the sort bar. Searching always
  shows matches regardless of folds.

Improved

• New sessions are written to disk immediately — they appear in the sessions list before
  the first message, and triggering "new session" on an already-empty session reuses it
  instead of minting another.
• Context trimming is now off by default (the full conversation is sent); the footer's
  ctx(%) readout tracks the same value as ctx and clamps at 100%.
• Default file-read limit lowered from 200 to 20 lines per read.
• Model picker hints moved to the bottom-right (CLI); the Max Context Window setting notes
  that 0 turns trimming off (VS Code).

Fixed

• Settings popup (VS Code) no longer loses an in-progress edit when closed by clicking
  outside or re-clicking the gear.
• CLI command output (e.g. /context-usage) is no longer hidden behind the TTFT/tok-s
  readout after the first completed turn.