lmstudio-agent-config

lmstudio-agent-config discovers models from LM Studio and generates local agent configuration for common developer clients.

Supported targets:

• VS Code + GitHub Copilot
• OpenCode
• Pi
• Codex

About

This tool reads LM Studio’s /api/v1/models endpoint, detects model capabilities, and produces client-specific configuration. You can print a model list or write config files directly for the supported targets, with optional filters for context length and capabilities.

Features

• Auto-discover LM Studio models from /api/v1/models.
• Detect tool calling, vision support, and context length.
• Generate configs for VS Code Copilot, OpenCode, Pi, and Codex.
• Preview changes with a focused diff and a confirmation prompt.
• Create dated backups before writing files.
• Filter by minimum context length, tools, and vision support.

How it works

• Queries LM Studio’s local server for available models.
• Builds target-specific configuration payloads.
• Either prints results or updates config files in place.

Safety behavior

When writing files, the tool is intentionally conservative. It prints a focused diff of only the changed lines, asks for confirmation (y/N), and only then writes the file. Before any modification, it creates a dated backup next to the original file (for example: settings.260215-0.backup.json), so you can roll back easily if needed.

Install

Run without install (PyPI):

uvx lmstudio-agent-config --help

Install with pip:

pip install lmstudio-agent-config

From source:

uv tool install .

Requirements

• LM Studio is running.
• The local server is enabled in LM Studio.
• The model listing endpoint is reachable (default: http://localhost:1234/api/v1/models).
• If LM Studio runs on another machine, use --base-url to point at that host and port.

Quick start

List discovered models:

lmstudio-agent-config

Write config for one target:

lmstudio-agent-config --settings code

Use a remote LM Studio server:

lmstudio-agent-config --base-url http://LMSTUDIO_HOST:1234/v1

Update all installed targets at once (skips missing default files):

lmstudio-agent-config --settings all

Usage

Use filters to control which models are included:

lmstudio-agent-config --min-context 32768
lmstudio-agent-config --tools
lmstudio-agent-config --no-tools
lmstudio-agent-config --vision
lmstudio-agent-config --no-vision
lmstudio-agent-config --min-context 32768 --tools --vision

Targets and default paths

Use --settings-path to override any default path.

Target	Flag	Default path
VS Code	--settings code	~/Library/Application Support/Code/User/settings.json
VS Code Insiders	--settings code-insiders	~/Library/Application Support/Code - Insiders/User/settings.json
OpenCode	--settings opencode	~/.opencode/opencode.json
Pi	--settings pi	~/.pi/agent/models.json
Codex	--settings codex	~/.codex/config.toml

Codex profiles

For Codex, the tool also generates one profile per discovered LM Studio model under profiles.*. Switch models with:

codex --profile lmstudio-your-model

Codex does not show these generated LM Studio profiles in a model picker, so use --profile explicitly when you want a non-default LM Studio model. The tool does not change top-level model_provider, so your default provider stays as-is.

Custom base URL

Use a custom LM Studio URL:

lmstudio-agent-config --base-url http://localhost:1234/v1

If LM Studio runs on another machine, you must point --base-url at that host and port.

Contributing

Issues and pull requests are welcome. For local development:

git clone https://github.com/alessandrobologna/lmstudio-agent-config.git
cd lmstudio-agent-config
uv sync --extra dev
uv run pytest
uv run ruff format --check
uv run ruff check
uv run ty check

If you want reproducible dev installs, use uv sync --extra dev --frozen.

License

MIT