Lemonade support

[antmikinka]

I need to update my agent workflow's work but wanted to show you what I have before you update again.
I do need to clean up the PR a bit remove a lot of stuff.
I do need to do some more manual testing.
Let me know your thoughts @Sportinger

Add Lemonade AI Provider (Optional Local AI)

Summary

Adds Lemonade Server as an optional local AI provider alongside OpenAI. Users can choose between:

Provider	Type	Privacy	Models	Speed
OpenAI	Cloud	Data sent to OpenAI	GPT-4, o3, etc.	Standard
Lemonade	Local	100% private	SmolLM3, Qwen2.5, etc.	Fast (local)

What This Adds

New Features (Optional)

• AI provider toggle in AIChatPanel (OpenAI ↔ Lemonade)
• Lemonade Server status indicator (online/offline)
• Model selection for Lemonade (SmolLM3, Qwen2.5, etc.)
• Fast fallback mode for quick responses

Files Added

File	Purpose
src/services/lemonadeProvider.ts	OpenAI-compatible provider for Lemonade Server
src/services/lemonadeService.ts	Server management, health checks, model listing

Files Modified

File	Changes
src/components/panels/AIChatPanel.tsx	Provider toggle UI, Lemonade status indicator
src/stores/settingsStore.ts	Lemonade settings (URL, model preference, enabled)
README.md	Lemonade badge and setup instructions

---

Master Branch Updates Included

This PR also brings in ~40 commits from master that lemonade-support was missing. These updates are already in master — this PR just ensures we're current.

Included Updates Summary

Feature	Status	Impact	Notes
Security hardening	✅ Required	Core	devBridgeAuth, fileAccessBroker — protects AI tool access from unauthorized file operations
AI Tool Policy	✅ Required	Core	Approval gating for destructive AI tools (delete, modify, export)
MediaBunny muxer	✅ Required	Core	Required for export pipeline — replaces mp4-muxer
Kie.ai integration	⚠️ Optional	Can Defer	Optional video generation provider — can move to separate branch
MatAnyone2	⚠️ Optional	Can Defer	Optional AI segmentation — can move to separate branch
Documentation audit	✅ Keep	None	Reference docs only — no runtime impact
Test suite	✅ Keep	None	Unit tests — no runtime impact

---

If You Want Minimal PR

We can split this into smaller PRs to make review easier:

Option A: Keep Together (Recommended)

Keep everything in this PR because:

• Security hardening is required for AI tools anyway
• MediaBunny is required for export to work
• Lemonade is a small addition (2 files)
• Kie.ai and MatAnyone2 are already in master

Option B: Split Into 3 PRs

PR #1: Security + Infrastructure (required foundation)

✅ Security hardening (devBridgeAuth, fileAccessBroker, redact)
✅ AI tool policy system
✅ MediaBunny muxer (export requirement)
✅ Documentation audit (reference only)
✅ Test suite

PR #2: Lemonade AI Provider (optional enhancement)

✅ lemonadeProvider.ts
✅ lemonadeService.ts
✅ AIChatPanel integration
✅ Settings store updates

PR #3: Optional AI Features (can defer indefinitely)

⚠️ Kie.ai video generation
⚠️ MatAnyone2 segmentation

---

Testing

Test	Status	Notes
Build passes (npm run build)	✅ Pass	No TypeScript errors
Lemonade Server connection	⏳ TODO	Requires Lemonade Server running locally
AIChatPanel provider toggle	⏳ TODO	Visual testing
OpenAI provider still works	⏳ TODO	Regression test
Export pipeline works	⏳ TODO	MediaBunny verification

---

Usage

For Users Who Want Lemonade

1. Install and run Lemonade Server
2. Open Settings → AI Features
3. Select "Lemonade" as AI provider
4. Choose model (SmolLM3, Qwen2.5, etc.)
5. Use AIChatPanel as normal

For Users Who Don't Want Lemonade

Nothing changes. The app works exactly as before:

• OpenAI provider is still default
• Lemonade is hidden until enabled
• No new dependencies affect core features

---

Future Updates

We can refine later:

Feature	Action	Priority
Kie.ai	Move to separate feature branch	Low
MatAnyone2	Move to separate feature branch	Low
Lemonade	Keep as optional local provider	High
Security	Keep in core (required for AI)	High
MediaBunny	Keep in core (required for export)	High

---

Screenshots

AIChatPanel with Provider Toggle

┌─────────────────────────────────────┐
│ AI Chat                    [⚙️]    │
├─────────────────────────────────────┤
│ Provider: [OpenAI ▼]  [🟢 Online]  │
│          or                                               │
│          [Lemonade ▼] [🟢 Online]  │
│                                     │
│ Model: [SmolLM3-3B-Instruct ▼]     │
│                                     │
│ ┌─────────────────────────────────┐ │
│ │ How can I help you today?       │ │
│ │                                 │ │
│ │ [Type your message...]          │ │
│ └─────────────────────────────────┘ │
│                                     │
│ [Generate] [Clear]                  │
└─────────────────────────────────────┘

Settings → AI Features

┌─────────────────────────────────────┐
│ AI Features                         │
├─────────────────────────────────────┤
│ Default Provider                    │
│ ○ OpenAI (cloud)                    │
│ ● Lemonade (local)                  │
│                                     │
│ Lemonade Server URL                 │
│ http://localhost:5000              │
│                                     │
│ Preferred Model                     │
│ [SmolLM3-3B-Instruct ▼]            │
│                                     │
│ [Save Changes]                      │
└─────────────────────────────────────┘

---

Related Issues

• Fixes: #XXX (Add Lemonade Server support)
• Related: #YYY (Local AI provider discussion)

---

Checklist

• Code follows project patterns
• Build passes without errors
• New files have appropriate exports
• Settings integrated into existing store
• Tested with running Lemonade Server
• Tested OpenAI provider still works
• Documentation updated

---

Reviewers: This PR brings Lemonade as an optional local AI provider. The main consideration is whether to include Kie.ai and MatAnyone2 in this PR or move them to separate branches. Recommend keeping them (they're already in master) but marking as optional features.

[Sportinger]

hi , thats a big one. will look into it asap. THANKS ! ;)

[antmikinka]

hi , thats a big one. will look into it asap. THANKS ! ;)

No problem. I mainly included the docs/phases/audits incase you wanted to look over for verification. Otherwise we can take them out casue it makes most of the 100k PR

[Sportinger]

Thanks for the PR.

Conceptually, I’m not sure Lemonade adds a fundamentally new capability here.

MasterSelects already exposes local AI control through the existing bridge / API layer, so a local LLM can already run
outside the app and drive the editor directly.

Right now that already works through:

• the dev server bridge: POST /api/ai-tools with the per-session Bearer token from .ai-bridge-token
• the Native Helper bridge: POST http://127.0.0.1:9877/api/ai-tools with the helper startup token
• the dev browser API: window.aiTools.execute(...)

So if the goal is simply "use a local LLM to control the editor", that path already exists.

Example dev-server call:

curl -X POST http://127.0.0.1:5173/api/ai-tools \

⚠ Heads up, you have less than 5% of your 5h limit left. Run /status for a breakdown.
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <token-from-.ai-bridge-token>" \
  -d '{"tool":"_status","args":{}}'

Example Native Helper call:

curl -X POST http://127.0.0.1:9877/api/ai-tools \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <startup-token>" \
  -d '{"tool":"_status","args":{}}'

This also seems to already cover external-agent / MCP-style workflows. I do not currently see a dedicated built-in MCP
server here, but the control surface for external tools is already present.

So to me Lemonade only really makes sense if the goal is specifically to provide a built-in local chat/provider inside
MasterSelects itself.

If the goal is just local LLM-driven editor control, the existing external bridge/API approach already seems like the
cleaner and more flexible architecture.

what do you think ?

[antmikinka]

Open to revise some work/plugins. There is NPU accel via Whisper stuff too so transcription is real fast
For future work, did you like looking over the agent docs, or should I just remove that before committing? I thought it would be interesting verification, but I can understand how it may be annoying for some.

PR Response: Lemonade Local AI (Short Version)

To: Reviewer
From: Lemonade PR Author (AMD intern, Lemonade team)

---

Full Disclosure + Context

I work on the Lemonade team as an AMD intern — I use Lemonade daily at home on my AMD AI PC with NPU acceleration. That's why I built this integration.

---

You're Right — Here's the Real Story

Your point is valid: The bridge already enables "local LLM controls editor."

"So to me Lemonade only really makes sense if the goal is specifically to provide a built-in local chat/provider inside MasterSelects itself."

Yes — exactly! That's the entire goal. This PR adds Lemonade as a built-in provider option (like OpenAI), not as a replacement for the bridge.

Your Implicit Point: "Why Not Just Document the Bridge?"

You're suggesting users could just use Lemonade via the bridge today — and you're right, they could.

Path	How It Works	What You Need
Bridge (works today)	Lemonade → Your Script → Bridge API → Editor	Write custom agent script
Built-in (this PR)	Lemonade → Settings Toggle → Chat Panel	Select from dropdown

Tradeoff:

• Bridge path = Requires writing custom agent script (power users)
• Built-in path = Toggle in Settings (everyone else)

We already have built-in OpenAI. This just adds local AI to the same dropdown.

What I should have been clearer about:

Lemonade vs Ollama — Technical Reality

Factor	Ollama	Lemonade
Backend	llama.cpp	llama.cpp (+ FastFlowLM for NPU)
Model format	GGUF	GGUF (same models work)
API	OpenAI-compatible at :11434	OpenAI-compatible at :8000
Model library	same	same
AMD NPU	Manual config	Auto via FastFlowLM
Built-in UI	No (requires bridge script)	Yes (this PR)
Claude Code support	✅ Via bridge	✅ Via bridge

Key point: Any GGUF model that works with Ollama works with Lemonade — same llama.cpp backend.

---

What This PR Actually Does

Adds Lemonade as a built-in option (2 files, optional):

• src/services/lemonadeProvider.ts — OpenAI-compatible provider
• src/services/lemonadeService.ts — Server health checks

Does NOT remove: Bridge architecture, OpenAI provider, or external agent support.

---

Honest Comparison: All Local AI Paths

Path	Setup	Best For
Ollama + Bridge	ollama run gemma3 + bridge script	Maximum model choice, power users
Lemonade + Bridge	lemonade-server run + built-in UI	AMD AI PC users (NPU auto-config), simple setup
Claude Code + Bridge	Install Claude Code + configure token	Complex reasoning tasks
OpenAI (cloud)	Enter API key	Best quality, no local hardware needed

---

My Proposal

1. Merge Lemonade as optional (2 files, low risk)
2. Document both Ollama and Lemonade fairly
3. Let users choose based on their hardware
4. Review in 3 months — if unused, can remove

---

TL;DR: You're right that bridge handles "local AI controls editor." Both Ollama and Lemonade work (same llama.cpp backend, GGUF models). Lemonade adds built-in UI + auto AMD NPU config. I use it daily on my AMD AI PC. Proposing we document both and let users choose.

---

Thanks for the thoughtful review — this clarity is valuable regardless of what we decide.

[Sportinger]

good point. will implenetn next week!

[antmikinka]

good point. will implenetn next week!

Fire! Let me know. I plan on making a lil YouTube video showing it working later this week.