LM Studio Guide

Use LM Studio v1 for the best local-server experience in 3sparks Chat. The native v1 provider uses LM Studio's /api/v1/* endpoints for rich model metadata, load/unload controls, stateful chats, response stats, reasoning controls, and MCP integrations.

If you want the phone itself to download and run a GGUF model without a desktop server, use Local Models.

Set up LM Studio

1. Install LM Studio from lmstudio.ai.
2. Download one or more models in LM Studio.
3. Start the LM Studio server.
4. Note the server address. From an iPhone or iPad, use your computer's LAN IP or Tailscale IP, not localhost.

Add the recommended v1 connection

1. Open 3sparks Chat and go to Connections.
2. Tap + and choose LM Studio v1.
3. Leave the name blank for a generated name, or enter your own.
4. Enter the base URL, usually http://YOUR_LOCAL_IP:1234.
5. Leave API Token blank unless LM Studio API tokens are enabled.
6. Tap Create Connection.
7. Tap Fetch Models, choose a default model, and tap Done.

Model loading and chat settings

For unloaded models, use Load to pick load options. Leave context length on Auto unless you know the model and computer can handle a specific value. Context length is separate from output token limits: context length controls how much LM Studio can consider, while output tokens control response length.

When a model exposes reasoning controls, 3sparks Chat shows supported reasoning options in chat settings. LM Studio response stats use the timing values reported by LM Studio.

MCP tools

LM Studio v1 connections can use MCP integrations. Configure saved MCP servers on the connection, then enable tools for chats that should use them. Tool restrictions configured in LM Studio are respected.

Keyboard behavior

The keyboard extension can use LM Studio v1 when the server is reachable. Keyboard requests are one-off transforms and do not create stateful LM Studio threads.

OpenAI-compatible fallback

Choose LM Studio - OpenAI compatible API only when you specifically need the /v1/chat/completions compatibility surface. Native LM Studio model management, MCP metadata, reasoning metadata, response stats, and request-level context length are not available through the compatibility provider.

Troubleshooting

• Use a LAN IP or Tailscale IP from iPhone and iPad; localhost points at the mobile device.
• Confirm LM Studio's server is running and listening on the expected port.
• If authentication is enabled, create an LM Studio API token and add it to the connection.
• If a model is unavailable, fetch models again or load the model in LM Studio.
• If a reasoning option errors, choose an option exposed by the selected model.