[santoshkumarradha]

Switch between regularly used chat models directly beside the composer: clicking the existing exchange glyph or pressing F3 advances through a saved rotation, updates the backend model and context window, and preserves the draft and cursor.

Settings → Models → quick switch models opens a searchable catalog. Enter adds/removes entries in selection order; selections persist across launches. The control is hidden with fewer than two entries, and F3 then opens its settings row. Rotations with three or more models show their position. Task rooms retain their own model controls.

Validation:

• make build and make pr-ready passed, including fresh config tests, all eight TUI shards, manual retrieval, and repository laws.
• Controlled integration coverage drives the actual mouse event through the TUI, real session, and HTTP backend; also covers persistence, context windows, draft preservation, glyph tiers, pointer alignment, and task isolation.
• Live OpenRouter verification after integrating current dev and restarting the application (tested code revision 1d75b7f69):

Chat turn	Requested and returned model	Result
1	deepseek/deepseek-v4.1-flash	HTTP 200, completed
2	deepseek/deepseek-v4-pro	HTTP 200, completed
3	deepseek/deepseek-v4.1-flash	HTTP 200, completed

All 32 recorded auxiliary calls used deepseek/deepseek-v4.1-flash. Clicking the exchange control also preserved an unsent draft.

A switch during a reply retains existing /model behavior: it may redirect the next request in that response. Removing and re-adding a saved model moves it to the end.

Screenshots are shared with the owner for review. This client cannot upload image attachments; the images are retained locally and no evidence files are committed.

Merge is pending the owner's approval.