learning_ai_common_plat/__LOCAL_LLMs/docs/DASHBOARD_REVIEW.md

Mission Control Dashboard — Bug & Improvement Review

Systematic code review of __LOCAL_LLMs/dashboard/ (6 source files, 1,395 lines) Last updated: Feb 19, 2026

---

File Inventory

File	Lines	Purpose
src/app/page.tsx	1,079	Main dashboard UI (single component)
src/app/globals.css	91	Design tokens, animations, base styles
src/app/layout.tsx	20	Root layout (metadata, dark mode)
src/app/api/ollama/route.ts	117	Ollama REST proxy (list, load, unload, pull, delete, show, generate)
src/app/api/ollama/stream/route.ts	38	Ollama streaming generate proxy (NDJSON)
src/app/api/whisper/route.ts	66	Whisper binary + GGML model discovery
src/app/api/system/route.ts	162	System info (chip, memory via vm_stat, disk, brew)

Stack: Next.js 16, React 19, TailwindCSS v4, Lucide icons, TypeScript

---

1. Bugs

• B1. Hardcoded machine specs in header — page.tsx:317 Subtitle reads Apple M4 Pro · 48 GB · {system?.platform} — should use system?.chip and formatBytes(system?.memory.total) dynamically so it works on any machine.

• B2. Pull model blocks UI — no progress feedback — api/ollama/route.ts:84-92 handlePull calls Ollama with stream: false. Large models (20+ GB) block for 30+ minutes. The Next.js API route will likely timeout. Must use stream: true and pipe progress events to the client. (Combined with F1.)

• B3. Dead code: non-streaming generate action — api/ollama/route.ts:69-82 The action === 'generate' handler is unused — UI only uses /api/ollama/stream. Remove or keep as fallback with a comment.

• B4. Escape key closes modal during active streaming — page.tsx:188-197 Global keydown handler calls setPromptModel(null) unconditionally. Backdrop click correctly checks !promptLoading. Escape should also respect promptLoading to prevent discarding an in-flight response.

• B5. Auto-refresh (15s) fires during streaming/pull — page.tsx:182-185 setInterval(fetchAll, 15000) runs unconditionally. During streaming this causes background churn and potential UI flicker. Should pause while promptLoading or pullLoading is true.

• B6. Toast ID collision on HMR remount — page.tsx:156-159 toastId.current resets to 0 on component remount during dev. Use Date.now() or crypto.randomUUID() for robust uniqueness.

• B7. vm_stat page size hardcoded — api/system/route.ts:103 Hardcoded 16384. Should parse from vm_stat's first line: "(page size of NNNNN bytes)" for portability.

• B8. Whisper models dir not configurable — api/whisper/route.ts:24 Hardcoded to ~/whisper-models. Should scan multiple known paths (/opt/homebrew/share/whisper-cpp/models/, ~/whisper-models, ~/.cache/whisper/) or accept WHISPER_MODELS_DIR env var.

• B9. No AbortController for streaming fetch — page.tsx:250-289 Closing the prompt modal doesn't cancel the underlying fetch. The reader.read() loop continues in the background wasting CPU/bandwidth until the model finishes generating.

• B10. Brew shows "Loading..." when array is empty — page.tsx:936-940 When system.brewPackages is [] (all uninstalled), displays "Loading..." instead of "No packages found". Needs to distinguish "still fetching" vs "fetched but empty".

• B11. Prompt text not cleared on close without send — page.tsx:951-957 Backdrop click clears promptText, but Escape handler (B4 fix) should also clear it. Otherwise stale text persists when re-opening.

---

2. Code Quality

• CQ1. Monolithic 1,079-line single component — page.tsx All interfaces, utilities, sub-components, and 900+ lines of JSX in one file. Extract to:

  • components/ — StatusDot, ProgressBar, ToastContainer, PromptModal, OllamaModelsPanel, SystemPanel, WhisperPanel, BrewPanel
  • lib/types.ts — interfaces (OllamaModel, SystemData, etc.)
  • lib/format.ts — formatBytes, formatUptime
  • lib/hooks.ts — useAutoRefresh, useToasts, useOllamaActions

• CQ2. Pervasive inline styles instead of CSS/Tailwind classes — page.tsx (100+ occurrences) Every style={{ color: 'var(--text-tertiary)' }} should be a utility class. Options: custom Tailwind theme mapping, or CSS utility classes in globals.css (e.g., .text-muted).

• CQ3. OLLAMA_URL duplicated — api/ollama/route.ts:3 + api/ollama/stream/route.ts:3 Same process.env.OLLAMA_URL || 'http://localhost:11434' in two files. Extract to lib/ollama-config.ts.

• CQ4. No React Error Boundary — page.tsx Unexpected API response shape crashes the entire dashboard. Add an error.tsx (Next.js App Router convention) for graceful recovery.

• CQ5. No loading skeleton / shimmer UI Initial load shows "..." placeholders. Skeleton cards would be more polished.

• CQ6. No TypeScript strict null checks in API responses API route handlers catch errors but return loosely typed JSON. Add Zod validation on the Ollama/system responses to prevent runtime surprises.

---

3. Features

• F1. Streaming pull with progress bar (fixes B2) Use Ollama stream: true for /api/pull. Create /api/ollama/pull/route.ts that pipes NDJSON progress. UI shows progress bar with completed/total bytes, speed, and ETA.

• F2. Model search/filter Search input above models list. Filter by name, family, quantization. Useful when 10+ models are installed.

• F3. Prompt history (localStorage) Store last 20 prompts with model name + timestamp. Dropdown in prompt modal to re-run previous prompts.

• F4. Chat mode (multi-turn conversation) Use Ollama /api/chat instead of /api/generate. Chat bubble layout with message history. System prompt input field.

• F5. Model comparison (side-by-side) Send same prompt to 2 models simultaneously. Display responses side-by-side with latency/quality comparison.

• F6. Token/s metrics after generation Parse eval_count and eval_duration from the final NDJSON chunk. Display tokens/second, total tokens, and latency in the response footer.

• F7. System resource sparklines (time-series) Ring buffer of memory/CPU snapshots (localStorage). Render mini sparkline charts in the System panel. Spot trends over time.

• F8. Ollama server logs viewer Read ~/.ollama/logs/ and display in a collapsible terminal-style panel. Filter by level. Auto-scroll.

• F9. Modelfile / template viewer The show action already fetches Modelfile, template, and system prompt. Display in a collapsible code block in expanded model details.

• F10. Dark/light theme toggle Add :root.light CSS variable overrides. Theme toggle with localStorage persistence. Current architecture supports this natively.

• F11. Keyboard shortcuts panel (? key) Show all shortcuts in a modal: ⌘+Enter (send), Esc (close), R (refresh), / (search models), ? (help).

• F12. Whisper transcription test Upload/record a short audio clip, transcribe locally via whisper-cli, display result with latency. Tests the full local STT pipeline.

• F13. Responsive mobile layout Better breakpoints for the 4-column stats row and 3-column main grid. Collapsible sidebar on mobile.

• F14. Model tags/labels (localStorage) User-defined tags (coding, fast, vision) with colored badges. Persisted in localStorage.

• F15. Extraction service integration panel Show extraction-service (port 4005) health status. Run test extractions against loaded Ollama models. Bridges dashboard to LysnrAI pipeline.

• F16. Auto-load preferred model Mark a model as "auto-load" (stored in localStorage). When Ollama is online but no models loaded, auto-load the preferred model.

---

4. Performance & Reliability

• P1. No request deduplication on Refresh — page.tsx:164-176 Rapid clicks on Refresh fire duplicate fetchAll() calls. Add a fetchingRef guard or disable the button during fetch (partially done for actionLoading but not for fetchAll).

• P2. Static cache never expires — api/system/route.ts:81-90 staticCache (chip, GPU, brew) lives forever in the server process. Brew package upgrades won't reflect. Add 5-minute TTL.

• P3. du -sk ~/.ollama/models on every refresh — api/system/route.ts:41 Traverses entire models directory every 15 seconds. Cache with 60-second TTL.

• P4. No fetch timeout on Ollama calls — api/ollama/route.ts:5-12 fetchOllama has no AbortSignal or timeout. If Ollama hangs, the dashboard hangs. Add 5-second timeout.

• P5. system_profiler slow on first load — api/system/route.ts:52-53 Takes ~2-3 seconds. Cached after first call, but first dashboard load waits. Consider eager background fetch on server start or return placeholder.

---

5. Security & Hardening

• S1. No input validation on model names — api/ollama/route.ts:50-51 model from request body passed directly to Ollama. Add regex validation: ^[a-zA-Z0-9._:/-]{1,256}$.

• S2. Shell command interpolation pattern — api/system/route.ts:67 execAsync(\brew list --versions ${pkg}`)— safe today (hardcoded targets) but fragile. UseexecFile('brew', ['list', '--versions', pkg])` for safety.

• S3. No CORS or auth (acceptable for local-only, documented) Any local process can call API routes. Fine for dev tool; document the assumption.

---

6. Implementation Tracker

Sprint 1 — Critical Bug Fixes (est. 1–2 hrs)

#	ID	Task	Effort	Commit
1	- [x] B4	Guard Escape key during streaming	5 min	2da67c2
2	- [x] B5	Pause auto-refresh during prompt/pull	10 min	2da67c2
3	- [x] B9	Add AbortController to streaming fetch	15 min	2da67c2
4	- [x] B1	Dynamic chip/RAM in header	5 min	2da67c2
5	- [x] B11	Clear prompt text on Escape close	5 min	2da67c2
6	- [x] P4	Add timeout to Ollama fetch calls	10 min	2da67c2
7	- [x] B3	Remove dead generate action (or document)	5 min	2da67c2
8	- [x] B6	Use Date.now() for toast IDs	2 min	2da67c2
9	- [x] B10	Fix brew "Loading..." vs "empty" state	5 min	2da67c2

Sprint 2 — Pull Progress + Metrics (est. 2–3 hrs)

#	ID	Task	Effort	Commit
10	- [x] B2+F1	Streaming pull with progress bar	60 min	2d9475b
11	- [x] F6	Display tokens/s after generation	30 min	2d9475b
12	- [x] B7	Parse vm_stat page size dynamically	10 min	2d9475b
13	- [x] B8	Multi-path whisper model discovery	15 min	2d9475b

Sprint 3 — Component Refactor (est. 2–3 hrs)

#	ID	Task	Effort	Commit
14	- [x] CQ1	Extract components into separate files	90 min	75a3cd0
15	- [x] CQ4	Add error.tsx Error Boundary	15 min	75a3cd0
16	- [x] CQ3	Shared ollama-config.ts	10 min	75a3cd0
17	- [x] CQ2	Consolidate inline styles → CSS classes	45 min	ed93a6f
18	- [x] S1	Add model name input validation	10 min	75a3cd0
19	- [x] S2	Replace exec → execFile for brew	10 min	75a3cd0

Sprint 4 — UX Enhancements (est. 3–4 hrs)

#	ID	Task	Effort	Commit
20	- [x] F3	Prompt history (localStorage)	45 min	9c2f5f3
21	- [x] F9	Modelfile viewer in expanded details	30 min	9c2f5f3
22	- [x] F4	Chat mode (multi-turn via /api/chat)	90 min	ed93a6f
23	- [x] F2	Model search/filter	30 min	9c2f5f3
24	- [x] F11	Keyboard shortcuts panel	20 min	9c2f5f3

Sprint 5 — Integration & Polish (est. 2–3 hrs)

#	ID	Task	Effort	Commit
25	- [x] F15	Extraction service panel	60 min	8bdd5ee
26	- [x] F12	Whisper transcription test	45 min	8bdd5ee
27	- [x] F7	System resource sparklines	45 min	8bdd5ee
28	- [x] CQ5	Loading skeleton UI	20 min	8bdd5ee
29	- [x] P1-P3	Request dedup + cache TTLs	30 min	b1fda3a
30	- [x] F16	Auto-load preferred model	20 min	ed93a6f

Deferred (nice-to-have)

ID	Task	Notes
- [x] F5	Model comparison (side-by-side)	8bdd5ee
- [x] F10	Dark/light theme toggle	ed93a6f
- [x] F13	Responsive mobile layout	8bdd5ee
- [x] F14	Model tags/labels	ed93a6f
- [x] CQ6	Zod validation on API responses	ed93a6f
- [x] F8	Ollama server logs viewer	8bdd5ee
- [x] S3	CORS / auth (documented)	8bdd5ee

---

7. Commit Log

Commits will be added here as work progresses.

#	Date	Commit	Sprint	Items Completed
1	Feb 19	2da67c2	Sprint 1	B1, B3, B4, B5, B6, B9, B10, B11, P4
2	Feb 19	2d9475b	Sprint 2	B2, B7, B8, F1, F6
3	Feb 19	75a3cd0	Sprint 3	CQ1, CQ3, CQ4, S1, S2
4	Feb 19	9c2f5f3	Sprint 4	F2, F3, F9, F11
5	Feb 19	b1fda3a	Sprint 5	P1, P2, P3
6	Feb 19	ed93a6f	Sprint 6	CQ2, CQ6, P5, F4, F10, F14, F16
7	Feb 19	8bdd5ee	Sprint 7	F5, F7, F8, F12, F13, F15, CQ5, S3

---

39 items total: 11 bugs, 6 code quality, 16 features, 5 performance, 3 security All 39 items completed across 7 sprints (9 code commits + doc updates) Actual total effort: ~8 hours across 7 sprints

---

8. Next Wave — Model Intelligence & Pre-Load Metrics

Proposed improvements focused on helping users make informed decisions before loading a model.

Tier A — Pre-Load Decision Metrics (est. 45 min)

ID	Feature	Description
N1	Estimated RAM per model	Approximate from disk size: Q4_K_M ≈ 1.2×disk in RAM. Show on every model card (e.g., ~22 GB RAM), not just running models.
N2	"Will it fit?" indicator	Compare estimated RAM vs system.memory.free + cached. Color-code: 🟢 Fits, 🟡 Tight (80–100%), 🔴 Won't fit. Show on Load button or as badge.
N3	Aggregate loaded model RAM	Sum VRAM of all running models. Display at top of models panel: "3 models loaded · 28.5 GB VRAM".

Tier B — Rich Model Metadata (est. 60 min)

ID	Feature	Description
N4	RAM budget bar	Horizontal stacked bar: `[OS+Apps	Model A (loaded)	Model B (loaded)	Free]`. Instant visual of memory headroom.
N5	Context window size	Fetch context_length from Ollama /api/show → model_info. Display on card (e.g., 128k ctx). Critical for knowing max prompt length.

Tier C — Model Intelligence Badges (est. 45 min)

ID	Feature	Description
N6	<think> warning badge	If model is DeepSeek R1 family, show ⚠️ badge: "Emits <think> traces — strip before JSON.parse". Prevents silent JSON failures.
N7	Vision model indicator	If model is multimodal (llava, qwen2.5vl), show 👁 badge. These need image input — text-only prompts are suboptimal.
N8	Architecture badge	Show model arch (llama, qwen2, phi3, deepseek2) as subtle pill on the card. Currently buried in expanded details.
N9	Sort/order models	Dropdown to sort by: name, size, parameters, running status, last modified. Currently uses Ollama's default order.
N10	Ollama version display	Call /api/version. Show in Ollama status card. Useful for debugging model compatibility.

Tier D — Runtime Metrics & UX (est. 30 min)

ID	Feature	Description
N11	Last known tok/s per model	Persist StreamMetrics.tokensPerSec in localStorage keyed by model. Show on card (e.g., ~45 tok/s). Compare speeds without re-benchmarking.
N12	Auto-unload countdown	Replace static Expires: 3:45 PM with live countdown: Unloads in 4m 32s. More actionable.
N13	Session stats per model	Track prompts sent + tokens generated per model in session. Show in expanded details.
N14	Delete confirmation + reclaim	Show "Delete qwen2.5-coder:32b? Reclaim 18.5 GB disk." before deleting. Currently no confirmation.
N15	Simultaneous load suggestions	Based on available RAM, suggest which models can be co-loaded. E.g., "Can co-load llama3.1:8b + qwen2.5-coder:32b (28 GB, 20 GB free)".

Implementation Plan

Sprint	Items	Focus	Effort
8	N1, N2, N3	Pre-load RAM estimates	~45 min
9	N4, N5	RAM bar + context window	~60 min
10	N6, N7, N8, N9, N10	Badges + sort + version	~45 min
11	N11, N12, N13, N14, N15	Runtime metrics + UX	~30 min