OpenAI-Compatible Endpoint for LM Studio

Route chat requests to any OpenAI-compatible API endpoint (local LAN, remote server) with system prompt presets from your LM Studio config presets. Select from your saved presets via dropdown, or override with custom text—perfect for image generation, video, coding, and other specialized prompts.

Features

• System Prompt Presets — Select from your LM Studio config presets (~/.cache/lm-studio/config-presets/*.preset.json) via dropdown. Automatically injects the preset's system prompt into upstream API requests.
• Custom System Prompt — Override any preset with your own text.
• Works with any OpenAI-compatible API endpoint (local LAN, self-hosted, or remote).
• Streaming chat responses.
• Tool call passthrough when the provider supports tools.
• Simple per-chat model selection.
• Optional debug logging for request/response diagnostics.

Requirements

• LM Studio installed, with the lms CLI available (LM Studio must be run at least once before lms works).

Quick Start (Local Development)

npm run dev runs lms dev, which starts the plugin dev server, verifies manifest.json, installs deps if needed, and rebuilds on changes.

Configuration

Configuration is done in LM Studio plugin settings.

Global (shared across chats)

• API Key: provider API key (leave empty if local endpoint doesn't require one)
• Base URL: provider base URL (e.g., http://192.168.x.x:port for local LAN)

Per-chat

• System Prompt Preset — Select an LM Studio config preset. Its system prompt is injected into the API request. Choose (none) to use the chat's built-in system prompt.
• Custom System Prompt — Override preset with custom text (leave empty to use selected preset).
• Model: provider model ID (e.g., Qwen/Qwen3.6-35B-A3B-FP8)
• Debug Logging: enable verbose request logging and status output

Usage

• Set API Key and Base URL in global settings (use your local API endpoint, e.g. http://localhost:port).
• Set the Model in the chat/plugin settings (e.g. Qwen/Qwen3.6-35B-A3B-FP8).
• Select a System Prompt Preset from the dropdown (or enter custom text).
• Start a chat.

If the model is missing, the plugin will return an error asking you to set it.

Troubleshooting

• 400 Provider returned error: the model ID may be invalid or the endpoint rejected the request. Enable Debug Logging to inspect the payload.
• Invalid or missing API key: verify the key in global settings (omit if your endpoint doesn't require authentication).
• Connection failed: verify the Base URL is reachable from LM Studio (check LAN IP, port, and that the server is running).

How It Works

System prompt presets are read from your LM Studio config preset files at:

Each preset file contains an llm.prediction.systemPrompt field. The plugin loads all presets at startup and presents them as a dropdown in the plugin settings. Selecting a preset causes its system prompt to replace the chat's system message in the API request sent to your endpoint.

Development Notes

• Source code lives in src/.
• Build output goes to dist/ (generated by TypeScript).
• Key files: src/generator.ts, src/config.ts, src/presets.ts, manifest.json.

Publishing

To publish your plugin to the LM Studio Hub:

Add -y to skip confirmation prompt. If not authenticated, run lms login first.

Security

Never hardcode API keys. Use the protected global config field instead.

License

ISC