Skip to main content
Local models in Kilo
Local runtime setup

Use LM Studio with Kilo Code for Local AI Coding

LM Studio is a practical choice when you want a desktop interface for discovering, downloading, configuring, and serving local models. Start its local model server, then connect Kilo to the provider on hardware you control.

Download LM Studio
Best for
Developers who want a visual model browser and runtime controls
Connection
Kilo local LM Studio provider
Model management
Download and configure models in the desktop app

Connect LM Studio to Kilo

  1. 1

    Install LM Studio

    Install LM Studio from its official site and let the app detect the available acceleration on your Mac, Windows, or Linux machine.

  2. 2

    Download a compatible model artifact

    Choose a coding-capable artifact and quantization that fits available memory. Keep enough headroom for the operating system, IDE, context, and tools.

  3. 3

    Load the model and start the local server

    Load the selected artifact in LM Studio and start its local server. Review the context and memory settings instead of assuming the advertised maximum will fit.

  4. 4

    Select LM Studio in Kilo and validate it

    Open Kilo provider settings, choose LM Studio, select the served model, and test a bounded coding request before starting a long agent workflow.

Before you connect

  • The model file, quantization, context length, and offload settings all affect memory use and latency.
  • A model that works for chat may not reliably follow tool schemas or complete multi-step coding tasks.
  • Local inference removes hosted token charges, but hardware, electricity, storage, and operator time still have costs.
  • Keep the local server bound to your machine unless you intentionally secure and operate it for remote access.

Troubleshooting

Kilo cannot reach LM Studio
Confirm that a model is loaded, the LM Studio local server is running, and Kilo is configured for the same endpoint. Check firewall and proxy settings if the runtime is remote.
The model fails to load
Choose a smaller artifact or stronger quantization, reduce context allocation, and close other memory-heavy applications.
Responses are slow
Check whether hardware acceleration and model offload are active. Smaller quantized models usually improve latency but may reduce coding or tool-use quality.
The agent loses context or misses tool calls
Verify the configured context window and use a model with demonstrated instruction-following and tool-use capability. Test a smaller task before increasing autonomy.

LM Studio and Kilo FAQ

Can I use LM Studio with Kilo Code in VS Code, JetBrains, and the CLI?

Yes. Kilo supports LM Studio as a local model provider across its local coding clients: the VS Code extension, JetBrains plugin, and CLI.

Should I use LM Studio or Ollama for coding?

Use LM Studio when you prefer a visual model browser and desktop controls. Use Ollama when you prefer command-line model management and a lightweight background service. Model quality and hardware fit matter more than the runtime alone.

Does LM Studio keep all Kilo activity offline?

LM Studio can keep model inference on your configured machine. Other networked Kilo features and integrations have separate data paths, so a fully offline deployment requires reviewing each enabled dependency.

Choose the right local coding model

Compare evidence-backed local model profiles by runtime, quantization, memory, and hardware tier before downloading a large artifact.

View local model recommendationsRead Kilo local model docs