LM Studio Integration

Got a desktop or laptop with a decent GPU? You can run a bigger model on it and let BlockVault talk to it over your home Wi-Fi. Your phone becomes a chat client, all the heavy lifting happens on your computer, and nothing leaves your network.

This is great if you want:

• More capable answers than what fits on a phone.
• Full privacy with your own hardware.
• No subscription, no cloud, no API keys.

How it works

┌──────────────┐    Wi-Fi / LAN    ┌──────────────────┐
│  BlockVault  │ ◄──── HTTP ─────► │    LM Studio     │
│  (phone)     │                   │  (Mac / Windows) │
└──────────────┘                   └──────────────────┘

1. LM Studio runs a local server on your computer.
2. You turn on "Serve on Local Network" so the phone can reach it.
3. In BlockVault you switch to LM Studio mode, paste the address, and pick a model.
4. The model runs on your computer's GPU/CPU. Responses stream back to your phone.

Step 1 — Install LM Studio

Download it from lmstudio.ai. It's free.

Platform	Notes
macOS	Apple Silicon (M1+) recommended — uses the Metal GPU automatically.
Windows	NVIDIA (CUDA) or AMD (Vulkan) GPU recommended. 8 GB RAM minimum, 16 GB+ ideal.

Step 2 — Download a model

Inside LM Studio, open the Discover tab and pick a model. A few tips:

• Tool-calling support matters. BlockVault uses tools (skills) to read your wallet, so pick a model trained for tool use. LM Studio shows a wrench icon or tools badge on supported models. The model picker in BlockVault also shows a tools tag next to compatible models.
• Match the model size to your RAM/VRAM. A 4–8B model is a good starting point on 8 GB machines. 12B+ models really benefit from 16 GB+ and a GPU with 10+ GB VRAM.
• Q4_K_M or Q8_0 quantizations are good defaults — smaller, faster, almost as good as full precision.
• Gemma 4 family pairs best with BlockVault, since the on-device assistant uses Gemma 4 too — behavior and tool-calling style stay consistent. Qwen 3 and similar tool-capable models also work well.

Step 3 — Start the server

1. Open the Developer tab in LM Studio (the <> icon).
2. Load the model you want to use.
3. Start the server (default port 1234).
4. Turn on "Serve on Local Network" — critical. Without it, only your computer itself can connect.

LM Studio will display the address, e.g. http://192.168.1.100:1234. Write it down.

Step 4 — Connect from BlockVault

1. Open the AI Assistant screen in BlockVault.
2. Switch the mode selector to LM Studio.
3. Paste the address (e.g. 192.168.1.100:1234) and tap the test button.
4. Pick a model from the list — you'll see size, context window, and a tools tag if it supports tool calling.
5. Tap Use This Model. BlockVault loads it with a generous context window (128k by default) and opens the chat.

You can switch back to on-device or to Delegate (remote GPU) at any time from the same screen.

Firewall reminders

• macOS: the first time, allow LM Studio in System Settings → Network → Firewall.
• Windows: allow LM Studio for both Private and Public networks when prompted.

Troubleshooting

Problem	Fix
"Could not connect"	Phone and computer must be on the same Wi-Fi; check IP and port.
Connection drops mid-chat	Computer went to sleep, Wi-Fi switched, or the server stopped — restart LM Studio.
"No models loaded"	Load a model in LM Studio before connecting.
Tool calls fail	Pick a model with tool-calling support (look for the wrench / tools badge).
Very slow responses	Try a smaller model or a lower quantization (e.g. Q4 instead of Q8).

Behind the scenes

BlockVault talks to LM Studio over two HTTP APIs on your LAN:

API	Used for
LM Studio native REST (/api/v1/)	Listing models, loading, unloading.
OpenAI-compatible (/v1/)	Streaming chat completions with tools.

No traffic leaves your local network.