Which remote servers does Off Grid support?

Any OpenAI-compatible server - Ollama, LM Studio, LocalAI, vLLM, and others. If it exposes a /v1/chat/completions endpoint, it works.

Does connecting to a remote server require internet?

No. Off Grid connects over your local WiFi network. No traffic goes to the internet. For access outside your home, use Tailscale.

Where are API keys stored?

In your device's system keychain via react-native-keychain. Never in plain storage.

Remote Servers - Connect Ollama, LM Studio, and LocalAI

Your phone can run impressive models locally, but your desktop or Mac can run much larger ones - Llama 3.1 70B, Mistral Large, DeepSeek, CodeLlama 34B.

Off Grid connects to any OpenAI-compatible server on your local network, giving you access to those models from your phone over WiFi. No internet required.

Supported servers

Server	Platform	Notes
Ollama	macOS, Linux, Windows	Most popular, easiest setup
LM Studio	macOS, Windows	Great UI, easy model management
LocalAI	Linux, Docker	Self-hosted, many model formats
vLLM	Linux	High-throughput, GPU-focused
Any OpenAI-compatible	Any	Needs `/v1/chat/completions` and `/v1/models`

Setting up Ollama

1. Install Ollama on your desktop:

# macOS
brew install ollama

# Linux
curl -fsSL https://ollama.ai/install.sh | sh

2. Allow remote connections (Ollama only listens on localhost by default):

# macOS/Linux - run Ollama with remote access
OLLAMA_HOST=0.0.0.0 ollama serve

# Or set permanently in ~/.zshrc / ~/.bashrc
export OLLAMA_HOST=0.0.0.0

3. Pull a model:

ollama pull llama3.1:8b
ollama pull qwen2.5:14b

4. Find your desktop’s local IP:

macOS: System Settings → Network → Wi-Fi → Details → IP address
Linux: ip addr show - look for your WiFi interface

Setting up LM Studio

Download and install LM Studio
Download a model in the app
Go to Local Server tab → click Start Server
Enable “Allow connections from network” in server settings
Note the IP and port shown (default port: 1234)

Connecting from Off Grid

Open Off Grid → Settings → Remote Servers
Tap Add Server
Enter the server URL:
- Ollama: http://192.168.1.42:11434
- LM Studio: http://192.168.1.42:1234
Add an API key if your server requires one (stored in system keychain)
Tap Test Connection → should show green
Tap Save

Off Grid will automatically discover all models available on the server via /v1/models.

Selecting a remote model

Open the model picker. Remote models appear under your server name. Tap one to make it active.

Off Grid streams responses via Server-Sent Events (SSE) in real time. Switching back to a local model is instant.

Vision and tool calling over remote servers

Off Grid detects vision and tool calling support from model name patterns. If the model name includes vision, vl, vlm, or similar, Off Grid enables the camera attachment. Tool calling is similarly detected.

For servers that support it (Ollama with compatible models, LM Studio), tool calling and vision both work without friction over the remote connection.

Access from outside your home with Tailscale

Tailscale creates a private VPN between your devices. Install it on both your desktop and phone, then use the Tailscale IP of your desktop as the server URL.

This gives you access to your home desktop’s models from anywhere - coffee shop, travel, office - without exposing anything to the public internet.

Security note

Off Grid warns you before connecting to a public internet endpoint (non-private IP range). For remote access, always use Tailscale or a similar private tunnel rather than exposing your server directly to the internet.