diff --git a/docs/roadmaps/not-started/LOCAL_AI_CONSOLIDATION_ROADMAP.md b/docs/roadmaps/not-started/LOCAL_AI_CONSOLIDATION_ROADMAP.md
index 68e212a6..522992f3 100644
--- a/docs/roadmaps/not-started/LOCAL_AI_CONSOLIDATION_ROADMAP.md
+++ b/docs/roadmaps/not-started/LOCAL_AI_CONSOLIDATION_ROADMAP.md
@@ -27,6 +27,127 @@ Both repos have a **chat workspace** that talks to Ollama, stores conversations,
 
 ---
 
+## Execution Checklist
+
+> **Track progress here.** Check each box as you go. Each `COMMIT` line is a commit point — verify before committing.
+> Update the `Status` header at the top of this file as you progress.
+
+### Phase 1 — Remove Workspace from LLM Lab (~2 hours)
+
+- [ ] **1.0** Pre-flight: `cd dashboard && pnpm test && pnpm run build` (establish green baseline)
+- [ ] **1.1** Delete `dashboard/src/app/(workspace)/` directory (22 files: 6 routes + 16 components)
+- [ ] **1.2** Delete `dashboard/src/app/api/ollama/title/` directory (1 workspace-only route)
+- [ ] **1.3** Delete workspace lib files: `db.ts`, `agents.ts`, `quick-actions.ts`, `scheduled-tasks.ts`, `cron.ts`, `migrate.ts`
+- [ ] **1.3a** Trim `types.ts`: move `Attachment` + `ModelDefaults` to shared section (lines 1–77), then delete everything from line 78 onward
+- [ ] **1.4** Delete workspace tests: `__tests__/agents.test.ts`, `__tests__/cron.test.ts`
+- [ ] **1.5** Create `dashboard/src/app/page.tsx` → `redirect('/mission-control')`
+- [ ] **1.6a** Remove npm deps from `package.json`: `idb`, `cron-parser`, `fuse.js`
+- [ ] **1.6b** Run `pnpm install` to update lockfile
+- [ ] **1.7** Verify: `pnpm run typecheck && pnpm test && pnpm run build`
+- [ ] COMMIT → `refactor(dashboard): remove chat workspace — LocalMemGPT is the chat surface`
+- [ ] **1.8a** Delete `e2e/chat.spec.ts`
+- [ ] **1.8b** Update `e2e/health.spec.ts` — remove `/c/test-id` test
+- [ ] **1.8c** Update `e2e/accessibility.spec.ts` — remove `/compare`, `/tts`, `/whisper` from routes
+- [ ] **1.8d** Update `e2e/visual-regression.spec.ts` — remove `workspace` and `compare` from ROUTES
+- [ ] **1.8e** Update `e2e/navigation.spec.ts` — `/` now redirects to `/mission-control`
+- [ ] **1.8f** Update `e2e/models.spec.ts` — `/` now redirects
+- [ ] COMMIT → `test(e2e): update specs after workspace removal`
+- [ ] **1.9** Update `AGENTS.md`, `README.md`, `CLAUDE.md`, `.windsurfrules`, `.cursorrules`
+- [ ] COMMIT → `docs: update agent docs after workspace removal`
+
+### Phase 2 — Cross-Links (~30 minutes)
+
+- [ ] **2.1** Add "Chat with models →" card/banner in Mission Control (LLM Lab repo, env: `LOCALMEMGPT_URL`)
+- [ ] **2.1a** Add `LOCALMEMGPT_URL` to `.env.example`
+- [ ] COMMIT → `feat(dashboard): add LocalMemGPT cross-link in Mission Control`
+- [ ] **2.2** Add "Model Management →" link in Settings panel (LocalMemGPT repo, env: `LOCALLLMLAB_URL`)
+- [ ] **2.2a** Add `LOCALLLMLAB_URL` to `.env.example`
+- [ ] COMMIT → `feat(web): add LLM Lab cross-link in settings`
+
+### Phase 3 (Optional) — Port Features to LocalMemGPT
+
+> Only after Phase 1+2 are complete and you've identified what's actually missed.
+
+- [ ] **3.1** Conversation Attachments (P1, ~1 day)
+- [ ] **3.2** Model Ratings (P1, ~2 hours)
+- [ ] **3.3** Projects / Folders (P1, ~4 hours)
+- [ ] **3.4** Custom Agents (P2, ~4 hours)
+- [ ] **3.5** Quick Actions (P2, ~6 hours)
+- [ ] **3.6** Voice Input — Whisper STT (P2, ~4 hours)
+- [ ] **3.7** Orchestrations (P3, ~1 day)
+- [ ] **3.8** Scheduled Tasks (P3, ~4 hours)
+
+### Phase 4 — DRY Extractions (~3–5 days, independent of Phases 1–3)
+
+- [ ] **4.1** Create `@bytelyst/ollama-client` package in `learning_ai_common_plat/packages/`
+- [ ] **4.1a** Migrate `learning_ai_local_memory_gpt/backend/src/lib/ollama.ts` → import from package
+- [ ] **4.1b** Simplify LLM Lab `api/ollama/*` routes + `ollama-config.ts` → import from package
+- [ ] **4.1c** Publish to Gitea registry
+- [ ] COMMIT → `feat(ollama-client): shared Ollama API client package`
+- [ ] **4.2** Add per-request helpers to `@bytelyst/fastify-sse` (startSSE, sendSSEData, endSSE)
+- [ ] **4.2a** Migrate LocalMemGPT `lib/sse-helpers.ts` → import from package
+- [ ] COMMIT → `feat(fastify-sse): add per-request SSE helpers`
+- [ ] **4.3** Create `@bytelyst/use-theme` package (configurable storage key + DOM strategy)
+- [ ] **4.3a** Migrate 6 web apps to import from package
+- [ ] COMMIT → `feat(use-theme): shared React theme toggle hook`
+- [ ] **4.4** Create `@bytelyst/use-keyboard-shortcuts` package (data-driven ShortcutDef[])
+- [ ] **4.4a** Migrate 5 web apps to import from package
+- [ ] COMMIT → `feat(use-keyboard-shortcuts): shared React keyboard shortcuts hook`
+
+---
+
+## Safety Matrix — Delete vs Keep
+
+> **Verified by import analysis.** No Mission Control or API route code imports any file in the DELETE column.
+
+### Files to DELETE (workspace-only)
+
+| File / Directory                                     | Verified By                                                                                                  |
+| ---------------------------------------------------- | ------------------------------------------------------------------------------------------------------------ |
+| `(workspace)/` — 6 routes + 16 components (22 files) | No imports from `(mission-control)/` or `api/`                                                               |
+| `api/ollama/title/` — 1 route                        | Only imported by `ConversationView.tsx` (deleted)                                                            |
+| `lib/db.ts` — IndexedDB client                       | Only imported by workspace components                                                                        |
+| `lib/agents.ts` — agent definitions                  | Only imported by workspace components                                                                        |
+| `lib/quick-actions.ts` — quick action defs           | Only imported by workspace components                                                                        |
+| `lib/scheduled-tasks.ts` — task scheduler            | Only imported by workspace components                                                                        |
+| `lib/cron.ts` — cron parser                          | Only imported by `TaskEditor.tsx`, `TaskRunner.tsx` (deleted)                                                |
+| `lib/migrate.ts` — IndexedDB migration               | Only imported by `db.ts` (deleted)                                                                           |
+| `types.ts` lines 78–118, 127–162, 170–221            | Workspace types: Conversation, Message, Agent, AgentTool, QuickAction, Project, ScheduledTask, Orchestration |
+| `__tests__/agents.test.ts`, `__tests__/cron.test.ts` | Test deleted lib files                                                                                       |
+| `e2e/chat.spec.ts`                                   | Tests workspace chat flow                                                                                    |
+| npm deps: `idb`, `cron-parser`, `fuse.js`            | Only used by deleted files                                                                                   |
+
+### Files to KEEP (Mission Control + shared infrastructure)
+
+| File / Directory                                                                                       | Used By (verified)                                              |
+| ------------------------------------------------------------------------------------------------------ | --------------------------------------------------------------- |
+| `(mission-control)/` — page.tsx + components                                                           | Core product — model ops, benchmarks, hardware                  |
+| `api/ollama/route.ts` — model list/pull/delete                                                         | Mission Control model management                                |
+| `api/ollama/pull/` — pull with progress                                                                | Mission Control model pull UI                                   |
+| `api/ollama/chat/` — chat proxy                                                                        | Mission Control model testing (page.tsx:719)                    |
+| `api/ollama/stream/` — streaming proxy                                                                 | Mission Control comparison/benchmarks (page.tsx:331,538)        |
+| `api/ollama/logs/` — log viewer                                                                        | Mission Control log viewer (page.tsx:297)                       |
+| `api/tts/` — TTS engine                                                                                | Mission Control TTS panel (page.tsx:179)                        |
+| `api/whisper/` — Whisper status                                                                        | Mission Control Whisper card (page.tsx:165)                     |
+| `api/whisper/transcribe/` — transcription test                                                         | Mission Control test button (page.tsx:313)                      |
+| `api/system/`, `system/memory/`, `system/exec/`                                                        | Mission Control hardware stats                                  |
+| `api/health/`, `api/extraction/`                                                                       | Infrastructure                                                  |
+| `lib/llm-router.ts`                                                                                    | `api/ollama/chat/route.ts`                                      |
+| `lib/ollama-config.ts`                                                                                 | All `api/ollama/*` routes                                       |
+| `lib/product-config.ts`                                                                                | Root layout                                                     |
+| `lib/router.ts`                                                                                        | `api/ollama/chat/route.ts` (imports Attachment + ModelDefaults) |
+| `lib/format.ts`                                                                                        | Mission Control page.tsx + `RamBudgetBar.tsx`                   |
+| `lib/use-theme.ts`                                                                                     | Mission Control page.tsx                                        |
+| `lib/use-keyboard-shortcuts.ts`                                                                        | Shared keyboard shortcuts                                       |
+| `types.ts` lines 1–77 + Attachment (119–126) + ModelDefaults (163–169)                                 | Mission Control types + `router.ts` dependency chain            |
+| `src/components/` — 6 shared components                                                                | Mission Control page.tsx imports all 6                          |
+| `__tests__/` — 5 test files (format, ollama-config, product-config, use-keyboard-shortcuts, use-theme) | Tests for kept lib files                                        |
+| `e2e/` — 6 specs (updated, not deleted)                                                                | Mission Control + navigation + accessibility                    |
+| `globals.css`                                                                                          | All shared styles (verified: no workspace-specific tokens)      |
+| `docs/`, `tts/`, `scripts/`, `experiments/`, `chat-history/`                                           | Non-dashboard assets                                            |
+
+---
+
 ## Review Findings
 
 > **Pass 1:** 13 bugs/gaps identified during systematic codebase verification (2026-03-29).
@@ -216,15 +337,15 @@ These are **optional ports** — only if they prove valuable:
 
 #### 1.1 Delete workspace route group
 
-Delete the entire `(workspace)/` directory:
+- [ ] Delete the entire `(workspace)/` directory:
 
 ```
-dashboard/src/app/(workspace)/        # 22 items — layout, pages, 16 components
+dashboard/src/app/(workspace)/        # 22 files — layout, 5 pages, 16 components
 ```
 
 #### 1.2 Delete workspace-specific API routes
 
-Delete **only** the route that is workspace-only:
+- [ ] Delete **only** the route that is workspace-only:
 
 ```
 dashboard/src/app/api/ollama/title/   # Auto-title generation (workspace-only)
@@ -241,7 +362,7 @@ dashboard/src/app/api/ollama/title/   # Auto-title generation (workspace-only)
 
 #### 1.3 Clean up lib files
 
-Delete workspace-only lib files:
+- [ ] Delete workspace-only lib files:
 
 ```
 dashboard/src/app/lib/db.ts              # IndexedDB (conversations, messages, agents, etc.)
@@ -252,10 +373,16 @@ dashboard/src/app/lib/cron.ts            # Cron parser
 dashboard/src/app/lib/migrate.ts         # v3→v4 migration
 ```
 
-Remove workspace types from `types.ts` (lines 78–221: Conversation, Message, Agent, QuickAction, Project, ScheduledTask, Orchestration). Keep lines 1–77 (OllamaModel, RunningModel, OllamaData, WhisperModel, WhisperData, SystemData, Toast, PullProgress, StreamMetrics). **[F9, R8] Also keep `Attachment` (lines 119–126) and `ModelDefaults` (lines 163–169)** — both are imported by `router.ts` (line 1: `import type { Attachment, ModelDefaults } from './types'`) which is used by `api/ollama/chat/route.ts` (shared).
+- [ ] Trim `types.ts` — **[F9, R8] do this carefully:**
+  1. Copy `Attachment` (lines 119–126) and `ModelDefaults` (lines 163–169) into the shared section (after line 77)
+  2. Delete everything from the original line 78 (`// --- v4 Workspace Types ---`) to end of file
+  3. **Why:** `router.ts` line 1 does `import type { Attachment, ModelDefaults } from './types'` → used by `api/ollama/chat/route.ts` (Mission Control)
+  4. **Verify:** `grep -r 'from.*types' dashboard/src/app/lib/router.ts` should still resolve
 
 #### 1.4 Delete workspace-specific tests
 
+- [ ] Delete workspace test files:
+
 ```
 dashboard/src/app/lib/__tests__/agents.test.ts
 dashboard/src/app/lib/__tests__/cron.test.ts
@@ -265,7 +392,7 @@ dashboard/src/app/lib/__tests__/cron.test.ts
 
 **[F3] CRITICAL:** The workspace route group `(workspace)/page.tsx` is the current root page (`/`). There is no separate `src/app/page.tsx`. After deleting the workspace, `GET /` returns 404.
 
-**Required:** Create `dashboard/src/app/page.tsx` that redirects to `/mission-control`:
+- [ ] Create `dashboard/src/app/page.tsx` that redirects to `/mission-control`:
 
 ```tsx
 import { redirect } from 'next/navigation';
@@ -274,8 +401,8 @@ export default function Home() {
 }
 ```
 
-- Verify `layout.tsx` root has no workspace-specific imports (confirmed: it only imports fonts, design tokens, product-config, and Providers — no workspace references)
-- Update any shared components that reference workspace routes
+- [ ] Verify `layout.tsx` root has no workspace-specific imports (confirmed: it only imports fonts, design tokens, product-config, and Providers — no workspace references)
+- [ ] Grep for any remaining imports of deleted files: `grep -r 'from.*\(workspace\)\|from.*db\|from.*agents\|from.*quick-actions\|from.*scheduled-tasks\|from.*cron\|from.*migrate' dashboard/src/`
 
 #### 1.6 ~~Update globals.css~~ — NO ACTION NEEDED [F8]
 
@@ -283,27 +410,38 @@ export default function Home() {
 
 #### 1.6a Remove workspace-only npm dependencies [F7]
 
-Remove from `dashboard/package.json`:
-
-- `idb` — IndexedDB client, only used by `db.ts` (workspace)
-- `cron-parser` — only used by `cron.ts` and `TaskEditor.tsx` / `TaskRunner.tsx` (workspace)
-- `fuse.js` — only used by `CommandPalette.tsx` (workspace)
+- [ ] Remove from `dashboard/package.json`:
+  - `idb` — IndexedDB client, only used by `db.ts` (workspace)
+  - `cron-parser` — only used by `cron.ts` and `TaskEditor.tsx` / `TaskRunner.tsx` (workspace)
+  - `fuse.js` — only used by `CommandPalette.tsx` (workspace)
+- [ ] Run `pnpm install` to update lockfile
 
 #### 1.7 Verify
 
+- [ ] Run and confirm all pass:
+
 ```bash
 cd dashboard && pnpm run typecheck && pnpm test && pnpm run build
 ```
 
-**Commit:** `refactor(dashboard): remove chat workspace — LocalMemGPT is the chat surface`
+- [ ] **COMMIT:** `refactor(dashboard): remove chat workspace — LocalMemGPT is the chat surface`
 
-#### 1.8 Update documentation
+#### 1.8 Update E2E specs
 
-- Update `AGENTS.md` — remove workspace references, update repo layout, adjust test count
-- Update `README.md` — remove chat workspace description, add note pointing to Local Memory GPT for chat
-- Update `CLAUDE.md`, `.windsurfrules`, `.cursorrules` — align with new scope
+- [ ] Delete `e2e/chat.spec.ts` — workspace chat flow
+- [ ] Update `e2e/health.spec.ts` — remove `Conversation Routes` test navigating to `/c/test-id`
+- [ ] Update `e2e/accessibility.spec.ts` — remove `/compare`, `/tts`, `/whisper` from routes array
+- [ ] Update `e2e/visual-regression.spec.ts` — remove `workspace` and `compare` from ROUTES
+- [ ] Update `e2e/navigation.spec.ts` — `/` now redirects to `/mission-control`
+- [ ] Update `e2e/models.spec.ts` — `/` now redirects
+- [ ] **COMMIT:** `test(e2e): update specs after workspace removal`
 
-**Commit:** `docs: update agent docs after workspace removal`
+#### 1.9 Update documentation
+
+- [ ] Update `AGENTS.md` — remove workspace references, update repo layout, adjust test count
+- [ ] Update `README.md` — remove chat workspace description, add note pointing to Local Memory GPT for chat
+- [ ] Update `CLAUDE.md`, `.windsurfrules`, `.cursorrules` — align with new scope
+- [ ] **COMMIT:** `docs: update agent docs after workspace removal`
 
 ---
 
@@ -315,22 +453,20 @@ cd dashboard && pnpm run typecheck && pnpm test && pnpm run build
 
 #### 2.1 Local LLM Lab → Local Memory GPT link
 
-Add a prominent link in Mission Control dashboard pointing to Local Memory GPT:
-
-- "Chat with your models →" card or banner
-- Links to `http://localhost:3070` (LocalMemGPT web default port)
-- Configurable via env var: `LOCALMEMGPT_URL`
-
-**Commit:** `feat(dashboard): add LocalMemGPT cross-link in Mission Control`
+- [ ] Add a prominent link in Mission Control dashboard pointing to Local Memory GPT:
+  - "Chat with your models →" card or banner
+  - Links to `http://localhost:3070` (LocalMemGPT web default port)
+  - Configurable via env var: `LOCALMEMGPT_URL`
+- [ ] Add `LOCALMEMGPT_URL` to `.env.example`
+- [ ] **COMMIT:** `feat(dashboard): add LocalMemGPT cross-link in Mission Control`
 
 #### 2.2 Local Memory GPT → Local LLM Lab link
 
-Add a link in Settings panel:
-
-- "Model Management →" pointing to `http://localhost:3000` (LLM Lab dashboard)
-- Configurable via env var: `LOCALLLMLAB_URL`
-
-**Commit:** `feat(web): add LLM Lab cross-link in settings`
+- [ ] Add a link in Settings panel:
+  - "Model Management →" pointing to `http://localhost:3000` (LLM Lab dashboard)
+  - Configurable via env var: `LOCALLLMLAB_URL`
+- [ ] Add `LOCALLLMLAB_URL` to `.env.example`
+- [ ] **COMMIT:** `feat(web): add LLM Lab cross-link in settings`
 
 ---
 
@@ -346,10 +482,11 @@ Only pursue these if the features are missed after Phase 1.
 
 LocalMemGPT currently has no attachment support. LLM Lab supports image/file/audio/url attachments.
 
-- Add `attachments` field to messages table (SQLite JSON column)
-- Update chat SSE handler to pass images to Ollama vision models
-- Update `MessageBubble.tsx` to render attachments
-- Update `ChatView.tsx` input to support file upload
+- [ ] Add `attachments` field to messages table (SQLite JSON column)
+- [ ] Update chat SSE handler to pass images to Ollama vision models
+- [ ] Update `MessageBubble.tsx` to render attachments
+- [ ] Update `ChatView.tsx` input to support file upload
+- [ ] **COMMIT:** `feat(chat): add attachment support for multimodal input`
 
 **Effort:** ~1 day
 
@@ -357,9 +494,10 @@ LocalMemGPT currently has no attachment support. LLM Lab supports image/file/aud
 
 LLM Lab has per-message thumbs up/down ratings with IndexedDB storage.
 
-- Add `rating` column to messages table
-- Add `PATCH /api/messages/:id/rate` endpoint
-- Update `MessageBubble.tsx` with rating buttons
+- [ ] Add `rating` column to messages table
+- [ ] Add `PATCH /api/messages/:id/rate` endpoint
+- [ ] Update `MessageBubble.tsx` with rating buttons
+- [ ] **COMMIT:** `feat(messages): add per-message model ratings`
 
 **Effort:** ~2 hours
 
@@ -367,10 +505,11 @@ LLM Lab has per-message thumbs up/down ratings with IndexedDB storage.
 
 LLM Lab groups conversations into projects. LocalMemGPT has no grouping.
 
-- Add `folders` table to SQLite
-- Add `folderId` column to conversations table
-- Add folder CRUD endpoints
-- Update Sidebar to show folder tree
+- [ ] Add `folders` table to SQLite
+- [ ] Add `folderId` column to conversations table
+- [ ] Add folder CRUD endpoints
+- [ ] Update Sidebar to show folder tree
+- [ ] **COMMIT:** `feat(conversations): add folder grouping`
 
 **Effort:** ~4 hours
 
@@ -378,10 +517,11 @@ LLM Lab groups conversations into projects. LocalMemGPT has no grouping.
 
 LLM Lab has 6 built-in agents + custom agent editor. LocalMemGPT supports per-conversation system prompts but no reusable agent definitions.
 
-- Add `agents` table to SQLite
-- Add agent CRUD endpoints
-- Add agent picker in new conversation flow
-- Port 6 built-in agent definitions
+- [ ] Add `agents` table to SQLite
+- [ ] Add agent CRUD endpoints
+- [ ] Add agent picker in new conversation flow
+- [ ] Port 6 built-in agent definitions
+- [ ] **COMMIT:** `feat(agents): add reusable agent definitions with built-in presets`
 
 **Effort:** ~4 hours
 
@@ -389,10 +529,11 @@ LLM Lab has 6 built-in agents + custom agent editor. LocalMemGPT supports per-co
 
 LLM Lab has 30+ built-in quick actions (code review, explain, summarize, etc.).
 
-- Add `quick_actions` table to SQLite
-- Add quick action CRUD + execute endpoints
-- Add Cmd+K command palette component
-- Port built-in quick action definitions
+- [ ] Add `quick_actions` table to SQLite
+- [ ] Add quick action CRUD + execute endpoints
+- [ ] Add Cmd+K command palette component
+- [ ] Port built-in quick action definitions
+- [ ] **COMMIT:** `feat(quick-actions): add command palette with 30+ built-in actions`
 
 **Effort:** ~6 hours
 
@@ -400,9 +541,10 @@ LLM Lab has 30+ built-in quick actions (code review, explain, summarize, etc.).
 
 LLM Lab has whisper.cpp integration for speech-to-text.
 
-- Add `POST /api/whisper/transcribe` endpoint (shell out to whisper binary)
-- Add microphone button in ChatView input bar
-- Detect whisper binary availability
+- [ ] Add `POST /api/whisper/transcribe` endpoint (shell out to whisper binary)
+- [ ] Add microphone button in ChatView input bar
+- [ ] Detect whisper binary availability
+- [ ] **COMMIT:** `feat(voice): add Whisper STT voice input`
 
 **Effort:** ~4 hours
 
@@ -410,9 +552,10 @@ LLM Lab has whisper.cpp integration for speech-to-text.
 
 LLM Lab supports chain/race/vote multi-model orchestrations. LocalMemGPT has compare (parallel) but not chain/vote.
 
-- Add `orchestrations` table
-- Add orchestration CRUD + execute endpoints
-- Add orchestration editor UI
+- [ ] Add `orchestrations` table
+- [ ] Add orchestration CRUD + execute endpoints
+- [ ] Add orchestration editor UI
+- [ ] **COMMIT:** `feat(orchestrations): add chain/race/vote multi-model pipelines`
 
 **Effort:** ~1 day
 
@@ -420,14 +563,257 @@ LLM Lab supports chain/race/vote multi-model orchestrations. LocalMemGPT has com
 
 LLM Lab has cron-based scheduled prompts. Very niche feature.
 
-- Add `scheduled_tasks` table
-- Add task scheduler (setInterval-based in backend)
-- Add task editor UI
+- [ ] Add `scheduled_tasks` table
+- [ ] Add task scheduler (setInterval-based in backend)
+- [ ] Add task editor UI
+- [ ] **COMMIT:** `feat(tasks): add cron-based scheduled prompts`
 
 **Effort:** ~4 hours
 
 ---
 
+### Phase 4 — DRY Refactoring & Common Platform Extractions
+
+> **Independent of Phases 1–3** — can be executed before, after, or in parallel.
+>
+> **Effort:** ~3–5 days total
+> **Risk:** Low-Medium — new shared packages, consumers migrate incrementally
+> **Repo:** `learning_ai_common_plat` (new packages) + both local AI repos + ecosystem-wide
+
+Code audit identified **7 extraction candidates** — duplicated patterns across the two local AI repos and the broader ecosystem.
+
+#### 4.1 `@bytelyst/ollama-client` — Ollama API Client (NEW PACKAGE)
+
+**Problem:** Both repos implement nearly identical Ollama API interaction code with duplicated NDJSON stream parsing.
+
+| Repo        | File                                           | Functions                                                                     | ~Lines |
+| ----------- | ---------------------------------------------- | ----------------------------------------------------------------------------- | ------ |
+| LocalMemGPT | `backend/src/lib/ollama.ts`                    | `listOllamaModels()`, `checkOllamaHealth()`, `streamChat()`, `getEmbedding()` | 117    |
+| LLM Lab     | `dashboard/src/app/api/ollama/route.ts`        | `fetchOllama()`, model list/pull/delete/show/load/unload                      | 128    |
+| LLM Lab     | `dashboard/src/app/api/ollama/stream/route.ts` | raw streaming proxy                                                           | 37     |
+| LLM Lab     | `dashboard/src/app/api/ollama/chat/route.ts`   | chat proxy with llm-router                                                    | 109    |
+| LLM Lab     | `dashboard/src/app/lib/ollama-config.ts`       | `OLLAMA_URL` env resolution + WSL2 gateway detection                          | 40     |
+
+**Shared NDJSON stream parsing pattern** (copy-pasted 5+ times across both repos):
+
+```ts
+const reader = res.body.getReader();
+const decoder = new TextDecoder();
+let buffer = '';
+while (true) {
+  const { done, value } = await reader.read();
+  if (done) break;
+  buffer += decoder.decode(value, { stream: true });
+  const lines = buffer.split('\n');
+  buffer = lines.pop() ?? '';
+  for (const line of lines) {
+    if (!line.trim()) continue;
+    try {
+      yield JSON.parse(line);
+    } catch {
+      /* skip */
+    }
+  }
+}
+```
+
+**Proposed package:** `packages/ollama-client/`
+
+```
+@bytelyst/ollama-client
+├── src/
+│   ├── config.ts          # resolveOllamaUrl(env) — OLLAMA_URL/OLLAMA_HOST + WSL2 detection
+│   ├── client.ts          # OllamaClient class: tags, ps, show, pull, load, unload, delete, version
+│   ├── stream.ts          # streamChat(), streamGenerate() — AsyncGenerator<OllamaStreamChunk>
+│   ├── embed.ts           # getEmbedding()
+│   ├── health.ts          # checkHealth() with timeout
+│   ├── ndjson.ts          # parseNdjsonStream() — reusable NDJSON async generator
+│   └── types.ts           # OllamaModel, OllamaChatMessage, OllamaStreamChunk, etc.
+└── package.json
+```
+
+**Consumers:**
+
+- [ ] `learning_ai_local_memory_gpt/backend` — replace `lib/ollama.ts` (117 lines deleted)
+- [ ] `learning_ai_local_llms/dashboard` — replace `lib/ollama-config.ts` + 4 API route files simplified
+- [ ] `@bytelyst/llm-router` — could consume `@bytelyst/ollama-client` for its Ollama provider
+
+**Effort:** ~1 day
+**Priority:** P0 — eliminates the largest duplication
+
+- [ ] **COMMIT:** `feat(ollama-client): shared Ollama API client package`
+
+---
+
+#### 4.2 Extend `@bytelyst/fastify-sse` — Per-Request SSE Utilities [R4]
+
+**Problem:** `@bytelyst/fastify-sse` provides `SSEHub` (broadcast to multiple clients) but NOT per-request SSE helpers. LocalMemGPT reinvented these as `sse-helpers.ts` (26 lines).
+
+| Existing                     | What it does                                                                           |
+| ---------------------------- | -------------------------------------------------------------------------------------- |
+| `@bytelyst/fastify-sse`      | `SSEHub` — multi-client broadcast pattern (hub/spoke)                                  |
+| LocalMemGPT `sse-helpers.ts` | `startSSE()`, `sendSSEData()`, `sendSSEEvent()`, `endSSE()` — single-request streaming |
+
+**Proposed:** Add per-request helpers to `@bytelyst/fastify-sse`:
+
+```ts
+// New exports in @bytelyst/fastify-sse
+export { startSSE, sendSSEData, sendSSEEvent, endSSE } from './per-request.js';
+```
+
+**Consumers:**
+
+- [ ] `learning_ai_local_memory_gpt/backend` — replace `lib/sse-helpers.ts` with import from package
+- [ ] `learning_ai_trails/backend` — already uses `@bytelyst/fastify-sse` hub, could use per-request too
+
+**Effort:** ~2 hours
+**Priority:** P1
+
+- [ ] **COMMIT:** `feat(fastify-sse): add per-request SSE helpers`
+
+---
+
+#### 4.3 `@bytelyst/use-theme` — React Theme Toggle Hook (NEW PACKAGE)
+
+**Problem:** Near-identical `useTheme()` hook duplicated across **6 web apps** — only the storage key and class-application strategy differ.
+
+| Repo        | File                                 | Storage Key | Lines |
+| ----------- | ------------------------------------ | ----------- | ----- |
+| LocalMemGPT | `web/src/lib/use-theme.ts`           | `lmg-theme` | 49    |
+| LLM Lab     | `dashboard/src/app/lib/use-theme.ts` | `llm-theme` | 48    |
+| ChronoMind  | `web/src/lib/use-theme.ts`           | `cm-theme`  | ~50   |
+| FlowMonk    | `web/src/lib/use-theme.ts`           | `fm-theme`  | ~50   |
+| NomGap      | `web/src/lib/use-theme.ts`           | `ng-theme`  | ~50   |
+| ActionTrail | `web/src/lib/use-theme.ts` [R5]      | `at-theme`  | ~50   |
+
+**Proposed package:** `packages/use-theme/`
+
+```ts
+export function useTheme(options?: { storageKey?: string; attribute?: 'class' | 'data-theme' }) {
+  // Configurable storage key (default: `${PRODUCT_ID}-theme`)
+  // Configurable DOM application (classList vs data-attribute)
+  // Returns { theme, setTheme, toggleTheme }
+}
+```
+
+**Effort:** ~2 hours
+**Priority:** P2 — saves ~300 total lines across 6 repos
+
+- [ ] **COMMIT:** `feat(use-theme): shared React theme toggle hook`
+
+---
+
+#### 4.4 `@bytelyst/use-keyboard-shortcuts` — Keyboard Shortcuts Hook (NEW PACKAGE)
+
+**Problem:** Near-identical `useKeyboardShortcuts()` hook duplicated across **5+ web apps**, each with slight variations in handler names.
+
+| Repo        | File                                                         |
+| ----------- | ------------------------------------------------------------ |
+| LocalMemGPT | `web/src/lib/use-keyboard-shortcuts.ts` (53 lines)           |
+| LLM Lab     | `dashboard/src/app/lib/use-keyboard-shortcuts.ts` (57 lines) |
+| ChronoMind  | `web/src/lib/use-keyboard-shortcuts.ts`                      |
+| ActionTrail | `web/src/lib/use-keyboard-shortcuts.ts`                      |
+| NoteLett    | `web/src/lib/use-keyboard-shortcuts.ts`                      |
+
+**Proposed package:** `packages/use-keyboard-shortcuts/`
+
+```ts
+export function useKeyboardShortcuts(shortcuts: ShortcutDef[]) {
+  // Each ShortcutDef: { key, meta?, shift?, handler, allowInInput? }
+  // Data-driven instead of hardcoded if/else chains
+}
+```
+
+**Effort:** ~3 hours
+**Priority:** P2 — saves ~250 lines, improves consistency
+
+- [ ] **COMMIT:** `feat(use-keyboard-shortcuts): shared React keyboard shortcuts hook`
+
+---
+
+#### 4.5 Ollama Model Types — Deduplicate into `@bytelyst/ollama-client` types
+
+**Problem:** `OllamaModel` interface is defined independently in both repos with slightly different shapes.
+
+| Repo                    | Definition                                                                                                              |
+| ----------------------- | ----------------------------------------------------------------------------------------------------------------------- |
+| LocalMemGPT `ollama.ts` | `{ name, size, digest, modified_at, details?: { parameter_size, quantization_level, family } }`                         |
+| LLM Lab `types.ts`      | `{ name, model, modified_at, size, digest, details: { families, family, format, parameter_size, quantization_level } }` |
+
+**Resolution:** Canonical `OllamaModel` type in `@bytelyst/ollama-client/types.ts` — matches Ollama's actual API response. Both repos import from there.
+
+**Effort:** included in 4.1
+**Priority:** P0 (part of 4.1)
+
+---
+
+#### 4.6 LLM Lab `format.ts` Utilities — Promote Reusable Functions
+
+**Problem:** `format.ts` (96 lines) contains both LLM Lab-specific functions AND general-purpose utilities:
+
+| Function                  | Scope                                                  | Extraction Target                                   |
+| ------------------------- | ------------------------------------------------------ | --------------------------------------------------- |
+| `formatBytes()`           | **Reusable** — any repo needs byte formatting          | `@bytelyst/ollama-client` or new `@bytelyst/format` |
+| `estimateRam()`           | LLM-infrastructure-specific                            | Keep in LLM Lab                                     |
+| `checkMemoryFit()`        | LLM-infrastructure-specific                            | Keep in LLM Lab                                     |
+| `getModelBadges()`        | LLM-infrastructure-specific                            | Keep in LLM Lab                                     |
+| `estimateTokens()`        | **Reusable** — any LLM app needs token estimation      | `@bytelyst/ollama-client`                           |
+| `getModelContextWindow()` | **Reusable** — any LLM app needs context window lookup | `@bytelyst/ollama-client`                           |
+| `formatUptime()`          | **Reusable** — monitoring/ops dashboards               | `@bytelyst/ollama-client`                           |
+
+**Effort:** ~1 hour (as part of 4.1)
+**Priority:** P1
+
+---
+
+#### 4.7 Client-Side SSE/NDJSON Stream Parser (Web)
+
+**Problem:** The client-side SSE parsing pattern is copy-pasted 3 times in LocalMemGPT's `web/src/lib/api.ts` (`streamChat`, `streamCompare`, plus a `parseSSEBody` in tests). LLM Lab has 3 more copies in `ConversationView.tsx`, `compare/page.tsx`, and `mission-control/page.tsx`. **[R6] Note:** After Phase 1, `ConversationView.tsx` and `compare/page.tsx` are deleted — LLM Lab reduces to 1 copy (Mission Control). Total post-Phase-1 copies: 4 (3 LocalMemGPT + 1 LLM Lab).
+
+**Proposed:** Export from `@bytelyst/ollama-client`:
+
+```ts
+// Client-side SSE stream consumer
+export function consumeSSEStream(
+  response: Response,
+  onData: (data: Record<string, unknown>) => void,
+  onDone: () => void
+): Promise<void>;
+
+// Client-side NDJSON stream consumer
+export function consumeNdjsonStream<T>(
+  response: Response,
+  onChunk: (chunk: T) => void
+): Promise<void>;
+```
+
+**Effort:** ~2 hours (as part of 4.1)
+**Priority:** P1 — eliminates 6 copy-pasted stream parsers
+
+---
+
+#### Phase 4 Summary
+
+| #   | Package                            | Action     | Lines Saved | Repos Affected          | Priority |
+| --- | ---------------------------------- | ---------- | ----------- | ----------------------- | -------- |
+| 4.1 | `@bytelyst/ollama-client`          | **NEW**    | ~350        | 2 local AI + llm-router | P0       |
+| 4.2 | `@bytelyst/fastify-sse`            | **EXTEND** | ~26         | LocalMemGPT + future    | P1       |
+| 4.3 | `@bytelyst/use-theme`              | **NEW**    | ~300        | 6 web apps              | P2       |
+| 4.4 | `@bytelyst/use-keyboard-shortcuts` | **NEW**    | ~250        | 5 web apps              | P2       |
+| 4.5 | (types in 4.1)                     | —          | included    | 2 local AI              | P0       |
+| 4.6 | (utils in 4.1)                     | —          | ~40         | 2 local AI              | P1       |
+| 4.7 | (client parsers in 4.1)            | —          | ~180        | 2 local AI              | P1       |
+
+**Total estimated savings:** ~1,150 lines of duplicated code across the ecosystem.
+
+**Recommended execution order:**
+
+1. **4.1** `@bytelyst/ollama-client` (biggest win, unblocks both repos)
+2. **4.2** Extend `@bytelyst/fastify-sse` (small, quick)
+3. **4.3 + 4.4** React hooks (lower priority, wider ecosystem impact — could batch with next DRY sweep)
+
+---
+
 ## File Inventory Summary
 
 ### Phase 1 Deletions (Local LLM Lab)
@@ -511,240 +897,6 @@ LLM Lab has cron-based scheduled prompts. Very niche feature.
 
 ---
 
-## Phase 4 — DRY Refactoring & Common Platform Extractions
-
-> **Note:** [R7] This phase was added as an addendum during the DRY audit (2026-03-29). It is independent of Phases 1-3 and can be executed before, after, or in parallel.
->
-> **Effort:** ~3–5 days total
-> **Risk:** Low-Medium — new shared packages, consumers migrate incrementally
-> **Repo:** `learning_ai_common_plat` (new packages) + both local AI repos + ecosystem-wide
-
-Code audit identified **7 extraction candidates** — duplicated patterns across the two local AI repos and the broader ecosystem.
-
-### 4.1 `@bytelyst/ollama-client` — Ollama API Client (NEW PACKAGE)
-
-**Problem:** Both repos implement nearly identical Ollama API interaction code with duplicated NDJSON stream parsing.
-
-| Repo        | File                                           | Functions                                                                     | ~Lines |
-| ----------- | ---------------------------------------------- | ----------------------------------------------------------------------------- | ------ |
-| LocalMemGPT | `backend/src/lib/ollama.ts`                    | `listOllamaModels()`, `checkOllamaHealth()`, `streamChat()`, `getEmbedding()` | 117    |
-| LLM Lab     | `dashboard/src/app/api/ollama/route.ts`        | `fetchOllama()`, model list/pull/delete/show/load/unload                      | 128    |
-| LLM Lab     | `dashboard/src/app/api/ollama/stream/route.ts` | raw streaming proxy                                                           | 37     |
-| LLM Lab     | `dashboard/src/app/api/ollama/chat/route.ts`   | chat proxy with llm-router                                                    | 109    |
-| LLM Lab     | `dashboard/src/app/lib/ollama-config.ts`       | `OLLAMA_URL` env resolution + WSL2 gateway detection                          | 40     |
-
-**Shared NDJSON stream parsing pattern** (copy-pasted 5+ times across both repos):
-
-```ts
-const reader = res.body.getReader();
-const decoder = new TextDecoder();
-let buffer = '';
-while (true) {
-  const { done, value } = await reader.read();
-  if (done) break;
-  buffer += decoder.decode(value, { stream: true });
-  const lines = buffer.split('\n');
-  buffer = lines.pop() ?? '';
-  for (const line of lines) {
-    if (!line.trim()) continue;
-    try {
-      yield JSON.parse(line);
-    } catch {
-      /* skip */
-    }
-  }
-}
-```
-
-**Proposed package:** `packages/ollama-client/`
-
-```
-@bytelyst/ollama-client
-├── src/
-│   ├── config.ts          # resolveOllamaUrl(env) — OLLAMA_URL/OLLAMA_HOST + WSL2 detection
-│   ├── client.ts          # OllamaClient class: tags, ps, show, pull, load, unload, delete, version
-│   ├── stream.ts          # streamChat(), streamGenerate() — AsyncGenerator<OllamaStreamChunk>
-│   ├── embed.ts           # getEmbedding()
-│   ├── health.ts          # checkHealth() with timeout
-│   ├── ndjson.ts          # parseNdjsonStream() — reusable NDJSON async generator
-│   └── types.ts           # OllamaModel, OllamaChatMessage, OllamaStreamChunk, etc.
-└── package.json
-```
-
-**Consumers:**
-
-- `learning_ai_local_memory_gpt/backend` — replace `lib/ollama.ts` (117 lines deleted)
-- `learning_ai_local_llms/dashboard` — replace `lib/ollama-config.ts` + 4 API route files simplified
-- `@bytelyst/llm-router` — could consume `@bytelyst/ollama-client` for its Ollama provider
-
-**Effort:** ~1 day
-**Priority:** P0 — eliminates the largest duplication
-
----
-
-### 4.2 Extend `@bytelyst/fastify-sse` — Per-Request SSE Utilities [R4]
-
-**Problem:** `@bytelyst/fastify-sse` provides `SSEHub` (broadcast to multiple clients) but NOT per-request SSE helpers. LocalMemGPT reinvented these as `sse-helpers.ts` (26 lines).
-
-| Existing                     | What it does                                                                           |
-| ---------------------------- | -------------------------------------------------------------------------------------- |
-| `@bytelyst/fastify-sse`      | `SSEHub` — multi-client broadcast pattern (hub/spoke)                                  |
-| LocalMemGPT `sse-helpers.ts` | `startSSE()`, `sendSSEData()`, `sendSSEEvent()`, `endSSE()` — single-request streaming |
-
-**Proposed:** Add per-request helpers to `@bytelyst/fastify-sse`:
-
-```ts
-// New exports in @bytelyst/fastify-sse
-export { startSSE, sendSSEData, sendSSEEvent, endSSE } from './per-request.js';
-```
-
-**Consumers:**
-
-- `learning_ai_local_memory_gpt/backend` — replace `lib/sse-helpers.ts` with import from package
-- `learning_ai_trails/backend` — already uses `@bytelyst/fastify-sse` hub, could use per-request too
-
-**Effort:** ~2 hours
-**Priority:** P1
-
----
-
-### 4.3 `@bytelyst/use-theme` — React Theme Toggle Hook (NEW PACKAGE)
-
-**Problem:** Near-identical `useTheme()` hook duplicated across **6 web apps** — only the storage key and class-application strategy differ.
-
-| Repo        | File                                 | Storage Key | Lines |
-| ----------- | ------------------------------------ | ----------- | ----- |
-| LocalMemGPT | `web/src/lib/use-theme.ts`           | `lmg-theme` | 49    |
-| LLM Lab     | `dashboard/src/app/lib/use-theme.ts` | `llm-theme` | 48    |
-| ChronoMind  | `web/src/lib/use-theme.ts`           | `cm-theme`  | ~50   |
-| FlowMonk    | `web/src/lib/use-theme.ts`           | `fm-theme`  | ~50   |
-| NomGap      | `web/src/lib/use-theme.ts`           | `ng-theme`  | ~50   |
-| ActionTrail | `web/src/lib/use-theme.ts` [R5]      | `at-theme`  | ~50   |
-
-**Proposed package:** `packages/use-theme/`
-
-```ts
-export function useTheme(options?: { storageKey?: string; attribute?: 'class' | 'data-theme' }) {
-  // Configurable storage key (default: `${PRODUCT_ID}-theme`)
-  // Configurable DOM application (classList vs data-attribute)
-  // Returns { theme, setTheme, toggleTheme }
-}
-```
-
-**Effort:** ~2 hours
-**Priority:** P2 — saves ~300 total lines across 6 repos
-
----
-
-### 4.4 `@bytelyst/use-keyboard-shortcuts` — Keyboard Shortcuts Hook (NEW PACKAGE)
-
-**Problem:** Near-identical `useKeyboardShortcuts()` hook duplicated across **5+ web apps**, each with slight variations in handler names.
-
-| Repo        | File                                                         |
-| ----------- | ------------------------------------------------------------ |
-| LocalMemGPT | `web/src/lib/use-keyboard-shortcuts.ts` (53 lines)           |
-| LLM Lab     | `dashboard/src/app/lib/use-keyboard-shortcuts.ts` (57 lines) |
-| ChronoMind  | `web/src/lib/use-keyboard-shortcuts.ts`                      |
-| ActionTrail | `web/src/lib/use-keyboard-shortcuts.ts`                      |
-| NoteLett    | `web/src/lib/use-keyboard-shortcuts.ts`                      |
-
-**Proposed package:** `packages/use-keyboard-shortcuts/`
-
-```ts
-export function useKeyboardShortcuts(shortcuts: ShortcutDef[]) {
-  // Each ShortcutDef: { key, meta?, shift?, handler, allowInInput? }
-  // Data-driven instead of hardcoded if/else chains
-}
-```
-
-**Effort:** ~3 hours
-**Priority:** P2 — saves ~250 lines, improves consistency
-
----
-
-### 4.5 Ollama Model Types — Deduplicate into `@bytelyst/ollama-client` types
-
-**Problem:** `OllamaModel` interface is defined independently in both repos with slightly different shapes.
-
-| Repo                    | Definition                                                                                                              |
-| ----------------------- | ----------------------------------------------------------------------------------------------------------------------- |
-| LocalMemGPT `ollama.ts` | `{ name, size, digest, modified_at, details?: { parameter_size, quantization_level, family } }`                         |
-| LLM Lab `types.ts`      | `{ name, model, modified_at, size, digest, details: { families, family, format, parameter_size, quantization_level } }` |
-
-**Resolution:** Canonical `OllamaModel` type in `@bytelyst/ollama-client/types.ts` — matches Ollama's actual API response. Both repos import from there.
-
-**Effort:** included in 4.1
-**Priority:** P0 (part of 4.1)
-
----
-
-### 4.6 LLM Lab `format.ts` Utilities — Promote Reusable Functions
-
-**Problem:** `format.ts` (96 lines) contains both LLM Lab-specific functions AND general-purpose utilities:
-
-| Function                  | Scope                                                  | Extraction Target                                   |
-| ------------------------- | ------------------------------------------------------ | --------------------------------------------------- |
-| `formatBytes()`           | **Reusable** — any repo needs byte formatting          | `@bytelyst/ollama-client` or new `@bytelyst/format` |
-| `estimateRam()`           | LLM-infrastructure-specific                            | Keep in LLM Lab                                     |
-| `checkMemoryFit()`        | LLM-infrastructure-specific                            | Keep in LLM Lab                                     |
-| `getModelBadges()`        | LLM-infrastructure-specific                            | Keep in LLM Lab                                     |
-| `estimateTokens()`        | **Reusable** — any LLM app needs token estimation      | `@bytelyst/ollama-client`                           |
-| `getModelContextWindow()` | **Reusable** — any LLM app needs context window lookup | `@bytelyst/ollama-client`                           |
-| `formatUptime()`          | **Reusable** — monitoring/ops dashboards               | `@bytelyst/ollama-client`                           |
-
-**Effort:** ~1 hour (as part of 4.1)
-**Priority:** P1
-
----
-
-### 4.7 Client-Side SSE/NDJSON Stream Parser (Web)
-
-**Problem:** The client-side SSE parsing pattern is copy-pasted 3 times in LocalMemGPT's `web/src/lib/api.ts` (`streamChat`, `streamCompare`, plus a `parseSSEBody` in tests). LLM Lab has 3 more copies in `ConversationView.tsx`, `compare/page.tsx`, and `mission-control/page.tsx`. **[R6] Note:** After Phase 1, `ConversationView.tsx` and `compare/page.tsx` are deleted — LLM Lab reduces to 1 copy (Mission Control). Total post-Phase-1 copies: 4 (3 LocalMemGPT + 1 LLM Lab).
-
-**Proposed:** Export from `@bytelyst/ollama-client`:
-
-```ts
-// Client-side SSE stream consumer
-export function consumeSSEStream(
-  response: Response,
-  onData: (data: Record<string, unknown>) => void,
-  onDone: () => void
-): Promise<void>;
-
-// Client-side NDJSON stream consumer
-export function consumeNdjsonStream<T>(
-  response: Response,
-  onChunk: (chunk: T) => void
-): Promise<void>;
-```
-
-**Effort:** ~2 hours (as part of 4.1)
-**Priority:** P1 — eliminates 6 copy-pasted stream parsers
-
----
-
-### Phase 4 Summary
-
-| #   | Package                            | Action     | Lines Saved | Repos Affected          | Priority |
-| --- | ---------------------------------- | ---------- | ----------- | ----------------------- | -------- |
-| 4.1 | `@bytelyst/ollama-client`          | **NEW**    | ~350        | 2 local AI + llm-router | P0       |
-| 4.2 | `@bytelyst/fastify-sse`            | **EXTEND** | ~26         | LocalMemGPT + future    | P1       |
-| 4.3 | `@bytelyst/use-theme`              | **NEW**    | ~300        | 6 web apps              | P2       |
-| 4.4 | `@bytelyst/use-keyboard-shortcuts` | **NEW**    | ~250        | 5 web apps              | P2       |
-| 4.5 | (types in 4.1)                     | —          | included    | 2 local AI              | P0       |
-| 4.6 | (utils in 4.1)                     | —          | ~40         | 2 local AI              | P1       |
-| 4.7 | (client parsers in 4.1)            | —          | ~180        | 2 local AI              | P1       |
-
-**Total estimated savings:** ~1,150 lines of duplicated code across the ecosystem.
-
-**Recommended execution order:**
-
-1. **4.1** `@bytelyst/ollama-client` (biggest win, unblocks both repos)
-2. **4.2** Extend `@bytelyst/fastify-sse` (small, quick)
-3. **4.3 + 4.4** React hooks (lower priority, wider ecosystem impact — could batch with next DRY sweep)
-
----
-
 ## Open Questions
 
 1. ~~**TTS/Whisper in Mission Control:**~~ **[F13] RESOLVED.** Mission Control uses `api/whisper` (status), `api/whisper/transcribe` (test), `api/tts` (engine status), `api/ollama/chat` (model testing), `api/ollama/stream` (benchmarks/compare), `api/ollama/logs` (log viewer). All 6 routes are kept.