Feature plan (living)
Product phases: ROADMAP.md · Docs: docs/
v0.3.0 slate shipped (KB + PDF, MCP Streamable HTTP + OAuth, TTS, edit message, notification actions, conversation summaries).
Shipped (do not re-implement)
| Area | Status | Notes |
|---|---|---|
| Streaming tool progress | Done | ToolProgress + EventChannel |
| Tasks / notes | Done | Services + screens + AI tools |
| Document extract/chunk | Done | + KB UI + Syncfusion PDF extract |
| Knowledge base → RAG | Done | KnowledgeBaseService + settings entry |
| MCP HTTP/SSE + stdio | Done | Legacy SSE still available |
| MCP Streamable HTTP + OAuth/PKCE | Done | Transport picker + token exchange UI |
| TTS speak response | Done | flutter_tts + Speak on bubbles |
| Edit user message + resend | Done | Truncate from turn and resend |
| Notification tap / actions | Done | Mirrors widget action IDs |
| Conversation summaries | Done | Extractive rolling summary → RAG |
| User memory overview | Done | Stored vs derived; promote; ask / inventory prompt |
| Assistant language | Done | Prefs match/en/de; system-prompt steering (UI stays English; full i18n later) |
| Message limits + stability guards | Done | Model-aware char caps; orchestrator pre-send validation |
| High context window + compact | Done | Gemma 4 KV 2048→4096 toggle; auto + manual compact of older turns |
| Adult mode | Done | Local prefs + system-prompt suffix; confirm dialog; still refuses illegal content |
| Settings JSON backup | Done | Export/import prefs, identity, MCP config, presets (no models) |
| Follow-up suggestions (bulb) | Done | 3 contextual chips + reroll above input |
| Response regeneration | Done | |
| Prompt presets | Done | |
| Parallel sessions | Done | Cap = 1 on Android |
| Screen capture | Done | On-demand JPEG; FGS on API 34+ |
| Inference stability | Done | No unload mid-stream; vision always-on |
Next up (priority order)
Plans: docs/superpowers/plans/2026-07-18-README.md
P0 — Hygiene
- Soak-test cleanup — Remove
AGENT_DBG/ ingest helpers after device verification (PR #6 / F1). Then bump 0.4.1.
P1 — UX / intelligence
- Conversation branching from a message — plan (ready; implement after greenlight)
- Share message / chat sheet — stub
- Continuous dictation / audio attach — stub
Free-RAM hard gate before Gemma 4 load- Heavy models / image gen — X8 Pro Max soak or LAN remote; not F1 (see models.md)
P2 — Platform
- Windows / Linux desktop hardening
- Optional wake word
P3 — Quality
- Raise unit/widget coverage toward 80%
- Integration tests for onboarding + model import
- Optional opt-in crash reporting
Explicit non-goals (for now)
- Re-adding
android.backup/or any duplicate Android tree - Shipping GGUF inference
- Starting stub
ModelServiceFGS “to keep model warm” (increases RAM pressure) - Continuous frame encode for screen capture
How to extend features
- Read AGENTS.md and docs/architecture.md
- Load nova-dev skill for setup/build/model flows
- Prefer small PRs; run
dart format,flutter analyze,flutter test - Update this file + ROADMAP when a checkbox-level item ships