Environment Variables

Create a .env.local file in the project root to override the defaults. All variables are optional — you can configure everything from the in-app Settings.

Legacy Variables (LM Studio)

These are the original variables. They still work but are superseded by the general provider variables below.

Variable	Default	Description
`LM_STUDIO_URL`	`http://localhost:1234/v1/chat/completions`	LM Studio chat completions endpoint.
`LM_STUDIO_MODEL`	`hy-mt1.5-7b/HY-MT1.5-7B-Q4_K_M.gguf`	Model identifier loaded in LM Studio.
`LM_STUDIO_TEMPERATURE`	`0.2`	Sampling temperature.

General Provider Variables

Server-side values used when the client doesn't provide per-request overrides.

Variable	Default	Description
`LLM_PROVIDER`	`lmstudio`	Provider to use: `lmstudio`, `ollama`, `openai`, `anthropic`, `gemini`, `openrouter`, `custom`.
`LLM_API_KEY`	—	API key for cloud providers (OpenAI, Anthropic, Gemini, OpenRouter).
`LLM_MODEL`	—	Model identifier (overrides per-provider default).
`LLM_TEMPERATURE`	`0.2`	Sampling temperature.

Provider-Specific URLs

Variable	Default	Provider
`LM_STUDIO_API_URL`	`http://localhost:1234`	LM Studio
`OLLAMA_API_URL`	`http://localhost:11434`	Ollama
`CUSTOM_API_URL`	—	Custom OpenAI-compatible

Example

# Local with LM Studio
LLM_PROVIDER=lmstudio

# Or cloud with OpenAI
# LLM_PROVIDER=openai
# LLM_API_KEY=sk-...
# LLM_MODEL=gpt-4o-mini

# Shared settings
LLM_TEMPERATURE=0.2

How environment variables work

The server-side API routes (/api/translate, /api/models, /api/test-connection) read these env vars as defaults. When a user configures their provider through the in-app Settings page, those values are stored in localStorage / sessionStorage and sent as per-request overrides — taking precedence over env vars.

Tips

URLs for LM Studio and Ollama should point to the server root (e.g. http://localhost:1234 — no /v1/...). The app builds the correct endpoint path automatically per provider.
Model must match the name shown in your provider exactly. Typos are the #1 cause of connection failures.
Temperature: 0.2 gives faithful translations. Bump it up for more creative outputs (at the cost of accuracy).
For runtime tweaks without file edits, use the in-app Settings.