Use with LM Studio

Short answer. LM Studio’s local API server listens on http://localhost:1234 by default. Start it from the Developer tab with the Start server switch, or from a terminal:

lms server start

Check that it answers and see the model identifiers it serves:

curl http://localhost:1234/v1/models

Atomic Agent ships a built-in LM Studio (local) preset for this address. Pick it in the desktop app or the terminal, choose a model, and the agent talks to LM Studio through its OpenAI-compatible endpoints. No API key is needed unless you turn on authentication in LM Studio.

LM Studio server cheat sheet

lms ships with LM Studio; run the app once before using it (lms docs).

Command	What it does
lms server start / lms server stop	Starts or stops the API server
lms server status	Shows whether the server runs and on which port
lms ls	Lists models on disk
lms ps	Lists models loaded in memory
lms load <model> --context-length=32768	Loads a model with a given context window
lms get qwen	Searches and downloads models

The port, network access and authentication live in the server settings: Server Port, Serve on Local Network and Require Authentication (LM Studio server settings).

Common LM Studio server errors

• Connection refused on port 1234. The server is off. Flip Start server in the Developer tab or run lms server start.
• Connection refused, but the server is on. You changed Server Port. Run lms server status to see the real port and use that number everywhere below.
• Model not found. The model id in the request does not match what /v1/models returns. Copy the id from that list.
• First answer takes a long time. With Just in Time Model Loading on, LM Studio loads the model on the first request. Load it ahead of time with lms load.
• 401 Unauthorized. Require Authentication is on and the client sent no token or a wrong one. See Authentication below.

Use LM Studio with Atomic Agent

The preset is an openai-compatible provider with base URL http://localhost:1234. Atomic Agent reads model ids from /v1/models and chats through /v1/chat/completions; both are listed in LM Studio’s OpenAI compatibility docs.

Desktop app

1. In LM Studio, download a model and start the server.

2. Open Atomic Agent. On the first-run screen, click Already run Ollama or LM Studio? Use it instead, or pick the Cloud models card, which also lists the model apps on your computer.

3. Choose LM Studio (local) and pick a model from the list LM Studio returns.

Already set up? Open Settings › Models › Cloud providers, click Add provider and choose LM Studio (local).

Terminal

1. Install Atomic Agent with curl -fsSL https://atomicagent.io/install | sh (Windows: irm https://atomicagent.io/install.ps1 | iex).

2. Run atomic-agent. In the first-run setup, choose Cloud models, then LM Studio (local), then a model.

3. Later, add it from the LLM tab: /llm, open Cloud, press n, pick LM Studio (local). Switch with /llm provider lmstudio and change models with /model.

Configure by hand

{
  "llm": {
    "activeTextProvider": "lmstudio",
    "providers": [
      { "id": "local-llama", "kind": "llama-server" },
      {
        "id": "lmstudio",
        "kind": "openai-compatible",
        "baseUrl": "http://localhost:1234",
        "apiKeyEnvVar": "LMSTUDIO_API_KEY",
        "defaultChatModel": "<model id from /v1/models>"
      }
    ]
  }
}

Keep the local-llama entry: it serves embeddings by default. defaultChatModel must be an id exactly as /v1/models returns it. An openai-compatible entry needs both baseUrl and defaultChatModel.

LM Studio on a different port or machine

The LM Studio (local) preset always saves http://localhost:1234. If you changed Server Port, or LM Studio runs on another computer with Serve on Local Network on, add an OpenAI-compatible provider instead (desktop: OpenAI-compatible API (custom base URL)) with the real address, for example http://192.168.1.20:1234, or edit baseUrl in the file.

If LM Studio requires authentication

With Require Authentication on, LM Studio expects a token in the Authorization header. Put the token in <stateDir>/.env (default ~/.atomic-agent/.env) as LMSTUDIO_API_KEY=..., the variable the preset names in apiKeyEnvVar, and restart the agent. With the variable unset, the agent sends no Authorization header at all.

Context length and tool calls

Atomic Agent reads only model ids from LM Studio, not each model’s context window. Set the window when you load the model (the load settings in the app, or lms load <model> --context-length=N). Agent sessions need room for tool definitions and history, so a few thousand tokens is not enough.

Tools go out in the OpenAI tools field, so the model must support tool calling. LM Studio describes this in its tool use docs. On Apple Silicon LM Studio can run both GGUF (llama.cpp) and MLX models (LM Studio docs); Atomic Agent does not care which engine serves the model, it only sees the API.

FAQ

What port does the LM Studio server use?

Port 1234 by default. You can change it in the server settings (Server Port). lms server status shows the port in use.

How do I start the LM Studio server without the app window?

Run lms server start. LM Studio also documents running headless as a service in its server docs.

Does Atomic Agent need an API key for LM Studio?

No, unless you enable Require Authentication. Then set LMSTUDIO_API_KEY in the agent’s .env.

Can Atomic Agent use MLX models from LM Studio?

Yes, as long as LM Studio serves them through its API. The agent sees an OpenAI-compatible server and a model id, whatever engine runs it.

Can I use Atomic Agent’s own model downloads instead?

Yes. The desktop app’s Local models route and the terminal’s atomic-agent models commands download GGUF models and run them on the agent’s built-in llama.cpp backend. See Use with llama.cpp.

Related

• Use with Ollama and Use with llama.cpp
• LM Studio vs Ollama
• Configuration