P1 — build the Quick Copilot summon window and full-chat handoff #86

New issue

Closed

opened 2026-08-12 14:42:07 +00:00 by lost-rob0t · 0 comments

lost-rob0t commented

2026-08-12 14:42:07 +00:00

(Migrated from github.com)

Motivation

Zara needs a fast Spotlight/Raycast-style interaction surface that can be summoned, used entirely from the keyboard, and expanded into the persistent chat without losing state.

Scope

Add a lightweight Quick Copilot PySide6 window backed by the same conversation/context models as #85.
Autofocus the composer on show.
Escape hides, Enter submits, Shift+Enter inserts newline.
Show compact assistant response, cancel/stop, model/provider indicator, voice/attachment affordances, and context chips.
Add one-action handoff into Full Chat while preserving the exact conversation ID, turn state and context attachments.
Recover sensible geometry on multi-monitor/display changes.
Keep show/hide fast; do not eagerly initialize heavy runtime components on every summon.

Acceptance criteria

Quick Copilot and Full Chat reference the same conversation model/service.
Expand/handoff does not duplicate messages or context.
Keyboard behavior matches the documented contract.
Repeated show/hide is stable and restores focus correctly.
Active generation can be cancelled from Quick Copilot.
Quick window works when invoked from tray even before global shortcuts are implemented.
Multi-monitor recovery prevents reopening entirely offscreen.

Testing requirements

Qt tests for focus, keyboard submit/newline/Escape, show/hide, cancellation and handoff.
Tests proving the same conversation ID/context survives handoff.
Geometry recovery tests.
Full repository/Nix suite.

Dependencies

#84 desktop shell.
#85 shared conversation/full chat.
Design: #81.

Explicit non-goals

No platform global-shortcut implementation in this issue.
No separate quick-mode LLM/runtime.
No always-on screen capture.
No duplicated persistence model.

## Motivation Zara needs a fast Spotlight/Raycast-style interaction surface that can be summoned, used entirely from the keyboard, and expanded into the persistent chat without losing state. ## Scope - Add a lightweight Quick Copilot PySide6 window backed by the same conversation/context models as #85. - Autofocus the composer on show. - `Escape` hides, `Enter` submits, `Shift+Enter` inserts newline. - Show compact assistant response, cancel/stop, model/provider indicator, voice/attachment affordances, and context chips. - Add one-action handoff into Full Chat while preserving the exact conversation ID, turn state and context attachments. - Recover sensible geometry on multi-monitor/display changes. - Keep show/hide fast; do not eagerly initialize heavy runtime components on every summon. ## Acceptance criteria - [ ] Quick Copilot and Full Chat reference the same conversation model/service. - [ ] Expand/handoff does not duplicate messages or context. - [ ] Keyboard behavior matches the documented contract. - [ ] Repeated show/hide is stable and restores focus correctly. - [ ] Active generation can be cancelled from Quick Copilot. - [ ] Quick window works when invoked from tray even before global shortcuts are implemented. - [ ] Multi-monitor recovery prevents reopening entirely offscreen. ## Testing requirements - Qt tests for focus, keyboard submit/newline/Escape, show/hide, cancellation and handoff. - Tests proving the same conversation ID/context survives handoff. - Geometry recovery tests. - Full repository/Nix suite. ## Dependencies - #84 desktop shell. - #85 shared conversation/full chat. - Design: #81. ## Explicit non-goals - No platform global-shortcut implementation in this issue. - No separate quick-mode LLM/runtime. - No always-on screen capture. - No duplicated persistence model.