ltcai 4.3.3 → 4.5.1

package/docs/V4_5_0_ONBOARDING_REPORT.md

@@ -0,0 +1,31 @@
+# v4.5.0 Onboarding Report
+
+Date: 2026-06-13
+
+## Restored Surface
+
+The desktop shell now shows a first-run setup guide unless the user dismisses it
+locally. It is also discoverable from the command palette as First-run Setup.
+
+## Covered Steps
+
+- Login: opens System -> Account.
+- Workspace Selection: opens System -> Workspaces.
+- Environment Analysis: opens Library -> Models and uses model recommendation
+  data from `/models/recommendations`.
+- Model Recommendation: surfaces the top recommended local model.
+- Model Installation: keeps runtime install/model download consent visible.
+- Model Validation: shows compatibility profile results after load.
+- Mode Selection: links to System -> Settings and preserves Basic/Advanced/Admin.
+- Brain Usage: opens the graph-first Brain surface.
+
+## Privacy Behavior
+
+No external model download, runtime install, or cloud call starts from
+onboarding. Downloads and installs require the visible Library consent checkbox.
+
+## Evidence
+
+Primary screenshot: `output/audits/v4.5.0-rc/screenshots/01-first-run-setup.png`
+
+Walkthrough GIF: `output/audits/v4.5.0-rc/gifs/v4.5.0-first-run-walkthrough.gif`

package/docs/V4_5_0_PRODUCT_EXPERIENCE_RECOVERY_REPORT.md

@@ -0,0 +1,49 @@
+# v4.5.0 Product Experience Recovery Report
+
+Date: 2026-06-13
+
+## Scope
+
+v4.5.0 restores the product experience on `main` after v4.4.0. It does not
+redesign `lattice_brain`, StorageEngine, FastAPI, Tauri, backup/restore, or
+portability architecture.
+
+## Restored Journey
+
+Before v4.5.0, account, workspace, setup, environment analysis, model setup, and
+mode selection existed but were scattered inside System/Library surfaces.
+
+After v4.5.0, the first-run path is visible from the app shell:
+
+1. Login
+2. Workspace Selection
+3. Environment Analysis
+4. Model Recommendation
+5. Model Installation
+6. Model Validation
+7. Mode Selection
+8. Brain Usage
+
+## Product Changes
+
+- Added a first-run guide in the desktop app shell with direct actions for
+  account, workspace, model setup, mode choice, and Brain entry.
+- Persisted the selected workspace across reloads so the chosen workspace is
+  used by API requests after restart.
+- Reworked Library Models around the existing streamed prepare/load path.
+- Kept model downloads and runtime installs behind explicit consent.
+- Added Gemma 4 MLX runtime metadata checks so 12B `gemma4_unified` shows
+  runtime update guidance while 26B A4B remains ready on the working path.
+- Reduced Basic-mode developer leakage while keeping Advanced/Admin detail.
+- Updated graph/search copy for readability, focus, and source clarity.
+
+## Evidence
+
+- Screenshots: `output/audits/v4.5.0-rc/screenshots/`
+- GIFs: `output/audits/v4.5.0-rc/gifs/`
+- Validation: `docs/V4_5_0_VALIDATION_REPORT.md`
+
+## No-Redesign Confirmation
+
+No storage migrations, archive format changes, FastAPI architecture changes,
+Tauri shell redesign, or Brain Core extraction changes were introduced.

package/docs/V4_5_0_VALIDATION_REPORT.md

@@ -0,0 +1,60 @@
+# v4.5.0 Validation Report
+
+Date: 2026-06-13
+
+## Status
+
+Local RC validation passed on macOS using Python 3.12.13
+(`/tmp/ltcai-v450-py312/bin/python`).
+
+## Required Matrix
+
+| Check | Result |
+| --- | --- |
+| Python compile | PASS: `npm run check:python` |
+| Ruff | PASS: `python -m ruff check .` |
+| Unit tests | PASS: `npm run test:unit` (`607 passed`) |
+| Integration tests | PASS: `npm run test:integration` against a live local server (`9 passed`, `1 skipped`) |
+| `lattice_brain` isolation tests | PASS: `tests/unit/test_lattice_brain_isolation.py` |
+| no `lattice_brain` -> `latticeai` import test | PASS: isolation import guard |
+| graph/search/ingestion tests | PASS: focused graph, search, and ingestion suite (`79 passed`) |
+| backup/restore tests | PASS: `tests/unit/test_kg_portability.py`, `tests/unit/test_v42_brain_storage.py` |
+| `.latticebrain` archive tests | PASS: encrypted archive round trip, tamper rejection, dry-run restore |
+| frontend lint/typecheck | PASS: `npm run lint`, `npm run typecheck` |
+| Playwright | PASS: `npm run test:visual` (`14 passed`) |
+| Tauri cargo check | PASS: `npm run desktop:tauri:check` |
+| Tauri DMG build | PASS: `npm run release:artifacts` |
+| release artifact validation | PASS: `npm run release:validate` |
+| wheel smoke | PASS: `scripts/wheel_smoke.py --wheel dist/ltcai-4.5.0-py3-none-any.whl` |
+| npm pack dry-run | PASS: `npm pack --dry-run` |
+| onboarding/model/Gemma/first-run validation | PASS: Playwright coverage plus screenshots/GIF evidence |
+
+## Notes
+
+- The first integration run was executed without the live local server and
+  failed with connection-refused setup errors. It was rerun against
+  `http://127.0.0.1:8899` and passed.
+- The Postgres live migration integration test remains skipped unless the
+  dedicated Postgres fixture is available.
+- Vite reports the existing large frontend bundle warning during production
+  builds.
+- Tauri/Rust reports an upstream future-incompatibility warning for
+  `block v0.1.6`.
+- SHA-256 artifact hashes were captured after the final local artifact rebuild
+  so this report can be packaged without creating self-referential hash churn.
+
+## Evidence
+
+- Screenshots: `output/audits/v4.5.0-rc/screenshots/`
+- Walkthrough GIF: `output/audits/v4.5.0-rc/gifs/v4.5.0-first-run-walkthrough.gif`
+- Validated artifacts:
+  - `dist/ltcai-4.5.0-py3-none-any.whl`
+  - `dist/ltcai-4.5.0.tar.gz`
+  - `ltcai-4.5.0.tgz`
+  - `dist/ltcai-4.5.0.vsix`
+  - `src-tauri/target/release/bundle/dmg/Lattice AI_4.5.0_aarch64.dmg`
+
+## External Publishing
+
+No PyPI, npm Registry, VS Code Marketplace, or Open VSX publish command is part
+of this RC validation.

package/docs/V4_5_1_GRAPH_EXPERIENCE_REPORT.md

@@ -0,0 +1,33 @@
+# v4.5.1 Graph Experience Report
+
+## Goal
+
+Knowledge exploration should feel like using a Digital Brain, not inspecting a
+database viewer.
+
+## Reimagined Entry Point
+
+Brain is now presented as Home. The default tab is Today, with the map still one
+click away. The first-viewport language now explains that Lattice turns added
+material into a living memory map and keeps every answer tied to its source.
+
+## Preserved Capability
+
+- Cytoscape graph explorer.
+- Semantic groups.
+- Search, focus, label modes, group collapse, and importance filtering.
+- Provenance coverage.
+- Hybrid search.
+- Portability and backup controls.
+
+## Presentation Changes
+
+- Page copy avoids endpoint/framework language in Calm mode.
+- Graph remains real and API-backed.
+- Existing graph route aliases still open the map directly.
+
+## Evidence
+
+- Graph UI: `frontend/src/pages/Brain.tsx`
+- Visual test: `knowledge graph renders a Cytoscape canvas and provenance coverage`
+- Related historical report: `docs/V4_5_0_GRAPH_UX_REPORT.md`

package/docs/V4_5_1_MODEL_EXPERIENCE_REPORT.md

@@ -0,0 +1,37 @@
+# v4.5.1 Model Experience Report
+
+## Goal
+
+Normal users should be able to install and use a local model without learning
+MLX, MLX-LM, MLX-VLM, Ollama, GGUF, or runtime internals.
+
+## Preserved Capability
+
+v4.5.1 keeps the v4.5.0 model recommendation and prepare/load stream:
+
+- Environment analysis.
+- Recommended model selection.
+- Explicit download/install consent.
+- Progress reporting.
+- Validation.
+- Load/ready state.
+- Compatibility recovery guidance.
+
+## Reimagined Presentation
+
+The model experience now sits under Library in the new product shell. The first
+session points users to "Pick a brain" and "Install locally" instead of runtime
+labels. Calm mode keeps implementation terms hidden while Deep/Admin retain
+diagnostics.
+
+## Non-Goals
+
+No model runtime behavior, download policy, local/cloud opt-in policy, or
+compatibility routing was changed in v4.5.1.
+
+## Evidence
+
+- Model UI: `frontend/src/pages/Library.tsx`
+- Model API client: `frontend/src/api/client.ts`
+- Related historical runtime report:
+  `docs/V4_5_0_MODEL_RUNTIME_UX_REPORT.md`

package/docs/V4_5_1_NAVIGATION_REPORT.md

@@ -0,0 +1,37 @@
+# v4.5.1 Navigation Report
+
+## New Navigation Model
+
+The primary navigation is now:
+
+- Home
+- Ask
+- Add
+- Automate
+- Library
+- Care
+
+This replaces the previous product taxonomy without removing compatibility
+routes. Existing paths such as `#/knowledge-graph`, `#/hybrid-search`,
+`#/files`, `#/agents`, `#/workspace-admin`, and admin aliases still route to
+the same capability modules.
+
+## Why This Model
+
+- Home is the user-facing Digital Brain entry point.
+- Ask is the clearest label for conversational work.
+- Add explains knowledge ingestion without technical vocabulary.
+- Automate explains agents/workflows as supervised outcomes.
+- Library remains clear for models, skills, and connections.
+- Care reframes account, backups, devices, and safety as maintenance.
+
+## Implementation
+
+- Route compatibility remains in `frontend/src/routes.ts`.
+- The visible dock and mobile drawer are implemented in `frontend/src/App.tsx`.
+- Visual tests now assert the new navigation labels and legacy route behavior.
+
+## Evidence
+
+- Tests: `tests/visual/v3.spec.js`
+- Code: `frontend/src/routes.ts`, `frontend/src/App.tsx`

package/docs/V4_5_1_ONBOARDING_REPORT.md

@@ -0,0 +1,29 @@
+# v4.5.1 Onboarding Report
+
+## Goal
+
+Onboarding should make the next action obvious to a first-time, non-technical
+user without documentation.
+
+## Reimagined Journey
+
+- Make it yours: account/profile setup.
+- Choose a space: personal or organization workspace selection.
+- Meet your Mac: environment analysis.
+- Pick a brain: model recommendation.
+- Install locally: consent-gated model/runtime setup.
+- Try a question: model validation through conversation.
+- Set the pace: Calm, Deep, or Admin mode.
+- Explore memory: open the knowledge map.
+
+## Behavior
+
+The journey reads real backend state from profile, workspace, models, and model
+recommendation APIs. The guide can be hidden through local storage, but no
+backend state is faked to mark setup complete.
+
+## Evidence
+
+- Primary component: `frontend/src/components/FirstRunGuide.tsx`
+- Desktop screenshot: `output/audits/v4.5.1-reimagining/screenshots/home-desktop.png`
+- Walkthrough GIF: `output/audits/v4.5.1-reimagining/gifs/v4.5.1-reimagining-walkthrough.gif`

package/docs/V4_5_1_PRODUCT_REIMAGINING_REPORT.md

@@ -0,0 +1,61 @@
+# v4.5.1 Product Reimagining Report
+
+Release date: 2026-06-13
+
+## Objective
+
+v4.5.1 treats the previous desktop product surface as non-authoritative. The
+goal was not to polish the v4.5.0 recovery UI; it was to relaunch Lattice AI as
+a premium local-first Digital Brain while preserving all capabilities,
+workflows, data contracts, storage, FastAPI APIs, Tauri shell behavior, Brain
+Core, model runtimes, backup/restore, and portability.
+
+## Product Decision
+
+The product is now organized around six user-facing rooms:
+
+- Home: the living shape of what Lattice knows.
+- Ask: thinking with remembered context.
+- Add: files, folders, and pages entering memory.
+- Automate: supervised goals, runs, workflows, approvals, hooks, and tools.
+- Library: models, skills, marketplace, and tool connections.
+- Care: account, spaces, backups, devices, settings, and admin safety.
+
+Legacy hash routes still resolve into the SPA, but they no longer dictate the
+visible product model.
+
+## What Changed
+
+- Replaced the fixed left rail with compact desktop chrome and a centered
+  navigation dock.
+- Added a command palette and responsive mobile drawer.
+- Added an ambient brain canvas so the first viewport reads as a thinking
+  environment rather than a dashboard.
+- Replaced the first-run checklist with a first-session journey.
+- Reworked route labels and page hero copy across Brain, Ask, Capture, Act,
+  Library, and System.
+- Rebuilt global styling around calmer surfaces, fixed responsive type sizes,
+  and restrained accents.
+
+## What Was Preserved
+
+- FastAPI and generated OpenAPI client.
+- Tauri desktop shell and sidecar behavior.
+- `lattice_brain` package boundary.
+- StorageEngine, SQLite default, optional PostgreSQL scale mode.
+- Backup/restore and `.latticebrain` portability.
+- Model recommendation, prepare/load stream, and consent-gated downloads.
+- Knowledge graph, hybrid search, chat, capture, workflow, agent, system, and
+  admin APIs.
+
+## Self Review
+
+The redesigned first viewport immediately reads as a different product:
+Lattice branding, Home/Ask/Add/Automate/Library/Care navigation, first-session
+journey, and premium desktop chrome replace the old dashboard hierarchy.
+
+## Evidence
+
+- Desktop screenshot: `output/audits/v4.5.1-reimagining/screenshots/home-desktop.png`
+- Mobile screenshot: `output/audits/v4.5.1-reimagining/screenshots/home-mobile.png`
+- Walkthrough GIF: `output/audits/v4.5.1-reimagining/gifs/v4.5.1-reimagining-walkthrough.gif`

package/docs/V4_5_1_RC_ARTIFACTS.md

@@ -0,0 +1,44 @@
+# v4.5.1 RC Artifact Manifest
+
+Release date: 2026-06-13
+
+## Screenshots
+
+- `output/audits/v4.5.1-reimagining/screenshots/home-desktop.png`
+- `output/audits/v4.5.1-reimagining/screenshots/home-mobile.png`
+- `output/audits/v4.5.1-reimagining/screenshots/memory-map.png`
+- `output/audits/v4.5.1-reimagining/screenshots/library-models.png`
+
+## GIFs
+
+- `output/audits/v4.5.1-reimagining/gifs/v4.5.1-reimagining-walkthrough.gif`
+
+## Release Artifacts
+
+- `dist/ltcai-4.5.1-py3-none-any.whl`
+- `dist/ltcai-4.5.1.tar.gz`
+- `ltcai-4.5.1.tgz`
+- `dist/ltcai-4.5.1.vsix`
+- `src-tauri/target/release/bundle/dmg/Lattice AI_4.5.1_aarch64.dmg`
+
+## SHA-256
+
+| Artifact | SHA-256 |
+| --- | --- |
+| `dist/ltcai-4.5.1-py3-none-any.whl` | `3d5ce1a0a85f7aba1f78587cd7e4a66c63dd5c03ddde7cee57624ec3f487899b` |
+| `dist/ltcai-4.5.1.tar.gz` | `62f0e05ff32554cf599b76678de3136bc02e5af4775144e7347182eed0fb4675` |
+| `ltcai-4.5.1.tgz` | `e755f40f87484d8a6e3f6bc95f48f0f78e1d0fcde3af8b14c709cf7fa71b2e4b` |
+| `dist/ltcai-4.5.1.vsix` | `3badc5915dc31425fa383d5946f78e0914497aa9e523cb7fbdc81a295b8f4a2f` |
+| `src-tauri/target/release/bundle/dmg/Lattice AI_4.5.1_aarch64.dmg` | `689ffc9553facf3987a5d57016dfd24fc3872f9c0810e28621f96b340ca38ce0` |
+
+## Validation
+
+`npm run release:validate` passed for the exact v4.5.1 artifact set. Historical
+files remain in `dist/` for prior releases, so all publish or upload steps must
+use exact v4.5.1 filenames only.
+
+## Notes
+
+The artifact list is exact-version only. Publishing remains out of scope for
+this RC unless the repository owner explicitly performs the manual registry
+publish steps.

package/docs/V4_5_1_UX_REPORT.md

@@ -0,0 +1,45 @@
+# v4.5.1 UX Report
+
+## UX Principle
+
+v4.5.1 designs for non-technical users who should understand what Lattice AI is,
+what to do next, how to add knowledge, how to use the brain, how to install a
+model, and how to automate work without reading the README.
+
+## Primary Flow
+
+The first-session journey is:
+
+1. Make it yours.
+2. Choose a space.
+3. Meet your Mac.
+4. Pick a brain.
+5. Install locally.
+6. Try a question.
+7. Set the pace.
+8. Explore memory.
+
+This replaces implementation-facing setup labels with user outcomes while still
+driving the same account, workspace, model recommendation, model setup, mode,
+and graph routes.
+
+## Screen Inventory Decision
+
+- Brain became Home, emphasizing the shape of work rather than a graph tool.
+- Capture became Add, emphasizing the act of feeding memory.
+- Act became Automate, emphasizing supervised work instead of an agent console.
+- System became Care, emphasizing trust, portability, and maintenance.
+- Ask and Library survived because their user intent is direct and legible.
+
+## Interaction Notes
+
+- Command palette remains available with Cmd/Ctrl+K.
+- Mode switching is visible as Calm, Deep, and Admin instead of Basic,
+  Advanced, and Admin in the first viewport.
+- Mobile navigation opens as a room chooser and preserves no-horizontal-overflow
+  behavior at 390px width.
+
+## Evidence
+
+- Playwright visual coverage: `tests/visual/v3.spec.js`
+- Screenshots: `output/audits/v4.5.1-reimagining/screenshots/`

package/docs/V4_5_1_VALIDATION_REPORT.md

@@ -0,0 +1,54 @@
+# v4.5.1 Validation Report
+
+Release date: 2026-06-13
+
+## Scope
+
+Validation covers the v4.5.1 product reimagining, version sync, frontend visual
+changes, release artifacts, and compatibility preservation.
+
+## Command Matrix
+
+| Check | Result |
+| --- | --- |
+| Python compile | PASS: `npm run check:python` compiled 775 modules |
+| Ruff | PASS: `node scripts/run_python.mjs -m ruff check .` |
+| Unit tests | PASS: `npm run test:unit` - 616 passed, 2 warnings |
+| Integration tests | PASS: `LTCAI_TEST_BASE_URL=http://127.0.0.1:4932 npm run test:integration` - 9 passed, 1 skipped |
+| Frontend lint | PASS: `npm run lint` |
+| TypeScript typecheck | PASS: `npm run typecheck` and `npm run typecheck:frontend` |
+| Playwright | PASS: `npm run test:visual` - 14 passed |
+| Tauri check/build | PASS: `npm run desktop:tauri:check`; PASS via `npm run release:artifacts` for the Tauri app and DMG build |
+| Release artifact validation | PASS: `npm run release:validate` |
+| Wheel smoke | PASS: `node scripts/run_python.mjs scripts/wheel_smoke.py --wheel dist/ltcai-4.5.1-py3-none-any.whl` |
+| npm pack dry-run | PASS: `npm pack --dry-run` |
+
+## Browser Evidence
+
+- Desktop screenshot: `output/audits/v4.5.1-reimagining/screenshots/home-desktop.png`
+- Mobile screenshot: `output/audits/v4.5.1-reimagining/screenshots/home-mobile.png`
+- Walkthrough GIF: `output/audits/v4.5.1-reimagining/gifs/v4.5.1-reimagining-walkthrough.gif`
+
+## RC Artifacts
+
+Expected exact-version artifacts:
+
+- `dist/ltcai-4.5.1-py3-none-any.whl`
+- `dist/ltcai-4.5.1.tar.gz`
+- `ltcai-4.5.1.tgz`
+- `dist/ltcai-4.5.1.vsix`
+- `src-tauri/target/release/bundle/dmg/Lattice AI_4.5.1_aarch64.dmg`
+
+## Artifact Hashes
+
+| Artifact | SHA-256 |
+| --- | --- |
+| `dist/ltcai-4.5.1-py3-none-any.whl` | `3d5ce1a0a85f7aba1f78587cd7e4a66c63dd5c03ddde7cee57624ec3f487899b` |
+| `dist/ltcai-4.5.1.tar.gz` | `62f0e05ff32554cf599b76678de3136bc02e5af4775144e7347182eed0fb4675` |
+| `ltcai-4.5.1.tgz` | `e755f40f87484d8a6e3f6bc95f48f0f78e1d0fcde3af8b14c709cf7fa71b2e4b` |
+| `dist/ltcai-4.5.1.vsix` | `3badc5915dc31425fa383d5946f78e0914497aa9e523cb7fbdc81a295b8f4a2f` |
+| `src-tauri/target/release/bundle/dmg/Lattice AI_4.5.1_aarch64.dmg` | `689ffc9553facf3987a5d57016dfd24fc3872f9c0810e28621f96b340ca38ce0` |
+
+`npm run release:validate` confirmed all exact-version RC artifacts are present.
+It also warned that historical artifacts remain in `dist/`, so publish commands
+must continue to use explicit v4.5.1 filenames rather than a `dist/*` glob.

package/docs/V4_5_1_VISUAL_DESIGN_REPORT.md

@@ -0,0 +1,30 @@
+# v4.5.1 Visual Design Report
+
+## Direction
+
+The visual language targets a calm premium desktop product rather than a
+developer/admin dashboard. The palette uses a carbon and warm-white base with
+jade, amber, violet, blue, and coral accents so the UI is not dominated by a
+single hue family.
+
+## System Changes
+
+- New sticky desktop chrome and centered dock.
+- Ambient brain canvas built from grid, line, and tile motifs.
+- Fixed responsive typography with media-query breakpoints, avoiding viewport
+  width font scaling.
+- Cards remain at 8px radius or less.
+- Shared buttons and badges were tightened for stable sizing and text fit.
+- Light theme retains the same hierarchy with warmer surfaces.
+
+## Accessibility And Layout
+
+- Primary navigation has an explicit `Primary navigation` label.
+- The command palette exposes a dialog label.
+- Mode switch has `Experience mode` semantics and pressed states.
+- Mobile layout was checked at 390x780 with zero horizontal overflow.
+
+## Evidence
+
+- Desktop visual pass: `output/audits/v4.5.1-reimagining/screenshots/home-desktop.png`
+- Mobile visual pass: `output/audits/v4.5.1-reimagining/screenshots/home-mobile.png`

package/docs/V4_DIGITAL_BRAIN_RECOVERY.md

@@ -5,21 +5,21 @@
 > completed analysis. **Update this file before ending any phase and before any
 > likely session/context/usage limit.**
 >
-> Last updated: 2026-06-13 — v4.3.3 Dead-Code Cleanup Release; Remaining Gaps remain empty
+> Last updated: 2026-06-13 — v4.5.1 Product Reimagining RC; Remaining Gaps remain empty
 
 ---
 
-## 0. RELEASE STATUS (v4.3.3)
-
-**v4.3.3 promotes the post-cleanup main tree after the independent dead-code,
-architecture, and runtime audit without redesigning the Digital Brain frontend,
-Brain Core, storage, or agent/workflow architecture.** Latest implementation
-milestone: v4.3.2 product polish plus post-audit cleanup, architecture
-documentation correction, Vercel/static-docs readiness, README badge
-restoration, and exact-current-main release artifacts.
-The v4.3.3 process creates a GitHub Release from locally validated artifacts.
-PyPI, npm Registry, VS Code Marketplace, Open VSX, and production deployments
-remain owner-only external publishing steps.
+## 0. RELEASE STATUS (v4.5.1)
+
+**v4.5.1 reimagines the desktop product surface after the v4.5.0 capability
+recovery without changing Brain Core, storage, API, model runtime, backup,
+portability, or agent/workflow behavior.** Current implementation milestone:
+new Home/Ask/Add/Automate/Library/Care navigation, compact desktop chrome,
+first-session journey, command palette, responsive mobile drawer, and refreshed
+Digital Brain visual system.
+The v4.5.1 RC process builds locally validated artifacts only. Tagging, GitHub
+Release creation, PyPI, npm Registry, VS Code Marketplace, Open VSX, and
+production deployments remain out of scope for this RC.
 Remaining implementation gaps: **none**.
 Owner-only blockers: pptx history rewrite (requires force-push/owner decision)
 and consent-gated production embedder provisioning (silent default download is
@@ -27,11 +27,11 @@ not permitted).
 
 ## Remaining Gaps
 
-None. v4.3.3 preserves the already-empty v4.3.2 gap list and promotes the
-post-cleanup tree: Brain graph exploration remains real and API-backed, raw
-product JSON dumps remain replaced with readable state, archive import/restore
+None. v4.5.1 preserves the already-empty v4.3.2/v4.3.3/v4.5.0 gap list and
+promotes the product-reimagining tree: Brain graph exploration remains real and
+API-backed, readable product state remains in place, archive import/restore
 flows remain exposed through existing APIs, desktop sidecar behavior remains
-validated, and exact v4.3.3 release artifacts are the release target.
+validated, and exact v4.5.1 RC artifacts are the validation target.
 Owner-only blockers above are intentionally not implementation gaps.
 
 ## 1. Program Charter (from the user's v4.0.0 directive)

package/docs/architecture.md

@@ -88,13 +88,17 @@ work on them.
 
 Local recommended models must be multimodal. The v2.2 local model workflow policy is:
 
-- macOS Apple Silicon: MLX-VLM first
+- macOS Apple Silicon: MLX-VLM first, with MLX-LM retained as a Gemma 4 text
+  fallback only for standard Gemma 4 metadata. Gemma 4 12B `gemma4_unified`
+  requires an MLX-VLM runtime that includes `mlx_vlm.models.gemma4_unified`.
 - Windows: llama.cpp multimodal path, with LM Studio as a user-friendly option
 - Linux: llama.cpp or vLLM multimodal path depending on GPU support
 - Ollama: kept as an option, not the default priority
 
-The removed path is the old text-only MLX-LM recommendation route. Low-spec
-machines use smaller or quantized multimodal models.
+The removed path is the old text-only MLX-LM recommendation lane for ordinary
+model selection. MLX-LM remains available as a targeted recovery path for
+standard Gemma 4 metadata, but it is not used for `gemma4_unified`.
+Low-spec machines use smaller or quantized multimodal models.
 
 ## Model Source Disclosure
 
@@ -140,7 +144,7 @@ Basic mode and advanced mode have the same feature access.
 | `latticeai/services/model_catalog.py` | Multimodal model catalog, source metadata, aliases |
 | `latticeai/services/model_recommendation.py` | Hardware-aware multimodal recommendation |
 | `latticeai/services/model_runtime.py` | Download, load, server, and model workflow orchestration |
-| `llm_router.py` | MLX-VLM and OpenAI-compatible model routing |
+| `llm_router.py` | MLX-VLM/MLX-LM and OpenAI-compatible model routing |
 | `knowledge_graph.py` | Graph storage, extraction, local folder knowledge graph context |
 | `latticeai/core/context_builder.py` | Graph context for generation |
 | `latticeai/core/workspace_os.py` | Workspace state, timeline, snapshots, durable context |