3.7 Local Free AI: LM Studio

LM Studio is a desktop tool to search, download, and run GGUF models with a GUI, then serve an OpenAI-compatible API via Local Server. With MiniTavern on your phone, you can chat using models running on your PC—no cloud API Key required.

---

Core configuration

When configuring LM Studio, prepare the following:

• Host IP: usually starts with 192.168 (e.g. 192.168.1.75)
• Base URL format: http://192.168.1.75:1234/v1

Important

• Use http (not https)
• Use port 1234 (LM Studio Local Server default; follow what the app shows)
• URL must be OpenAI-compatible, with /v1 at the end

Example: http://192.168.1.75:1234/v1 (where 192.168.1.75 is the LAN IP shown in LM Studio)

---

Why LM Studio?

• GUI: search, download, and load models with little command-line work
• Free locally: local inference does not require a cloud API Key
• Privacy: conversations go over your LAN to your PC for processing
• OpenAI-compatible: works with MiniTavern Other LLM
• Wide model choice: rich GGUF catalog with many quantization options

---

Requirements

• A computer (macOS example; Windows/Linux work too)
• Phone and PC on the same Wi-Fi
• PC stays on with LM Studio Local Server running
• Enough RAM/VRAM (7B+ recommended for roleplay)

---

Steps (macOS example)

Preparation: install LM Studio and download a model

1. Download and install LM Studio from https://lmstudio.ai/, then open the app.
2. Open Discover (or the model library), search for and download a moderately sized instruct model for testing—for example a Qwen2.5-7B-Instruct GGUF build (exact name varies in the catalog). Wait for the download to finish.

The first model download uses significant disk space—plan ahead.

Step 1: start Local Server and allow LAN access

Complete these two checkpoints before continuing:

• Checkpoint 1: turn on the LM Studio local server switch. When running, the UI should show Status: Running.
• Checkpoint 2: in Server Settings, enable Serve on Local Network so phones and other devices on the same Wi-Fi can reach LM Studio on your PC.

Why enable LAN access?

• MiniTavern runs on your phone and must reach LM Studio on your PC over the LAN
• By default, only 127.0.0.1 may be allowed
• After enabling, phones on the same Wi-Fi can use http://PC_IP:1234/v1

Step 2: confirm the LAN address and load a model

Complete these two checkpoints:

• Checkpoint 3: check the LAN address LM Studio displays—for example http://192.168.1.75:1234 in the screenshot. This is the prefix for the MiniTavern Base URL (you still append /v1 at the end).
• Checkpoint 4: use the Load Model button (top right) to select and load a downloaded local model. After loading, running model info appears below.

Optional self-test in a terminal:

curl http://127.0.0.1:1234/v1/models

JSON with model ids means Local Server is ready.

Step 3: configure MiniTavern

1. Open MiniTavern → Settings → LLM settings → Configure LLM (or AI provider).
2. Under API Provider, choose Other LLM.
3. In Base URL, enter the address from Checkpoint 3 and append /v1.

For example, if Checkpoint 3 shows http://192.168.1.75:1234, set Base URL to:

http://192.168.1.75:1234/v1

4. Tap Model, select the model currently loaded in LM Studio, and run a connection test.

Important

• Use http (not https)
• Port must be 1234 (or what LM Studio shows)
• URL must end with /v1 (OpenAI-compatible path)

API Key: lm-studio or any placeholder (local servers usually do not validate keys).

If connection fails:

• Confirm the model is loaded and Local Server is running (Status: Running)
• Confirm Serve on Local Network is enabled
• Confirm Base URL matches the LAN address in LM Studio and ends with /v1
• Confirm phone and PC are on the same Wi-Fi
• Check whether the firewall blocks port 1234

After a successful test, save and start chatting with a character card.

---

FAQ

Q: Why use the /v1 suffix?

A: This is the OpenAI-compatible request format. LM Studio has its own API format, but OpenAI-style requests are more common in the ecosystem. MiniTavern Other LLM connects this way too. If LM Studio’s /api/v1 format causes issues, switch to this guide’s http://IP:1234/v1 approach.

Q: Empty model list?

A:

• Confirm the model is Loaded and Local Server is running;
• In a browser on the PC, open http://127.0.0.1:1234/v1/models and check for a response;
• Confirm the MiniTavern URL includes /v1.

Q: Poor reply quality?

A:

• Try a larger model or higher quantization (7B+);
• Adjust context length in LM Studio;
• Try different character cards and presets.

---

Related links

• LM Studio
• 3.3 AI providers
• 2.2 API settings (advanced)