Configuration

Local environment file

For local development and end-to-end tests, copy the template from the repository root:

cp .env.example .env

Fill in values for the backends you use. .env is gitignored.

Common variables

The root .env.example is the source of truth. It includes:

Area	Examples
OpenAI-compatible	OPENAI_API_KEY, OPENAI_BASE_URL, OPENAI_MODEL, OPENAI_EMBEDDING_MODEL, OPENAI_TRANSCRIPTION_MODEL, …
OpenRouter	OPENROUTER_API_KEY, OPENROUTER_BASE_URL, OPENROUTER_MODEL, optional app headers
Gemini	GEMINI_API_KEY, GEMINI_BASE_URL, GEMINI_MODEL, GEMINI_EMBEDDING_MODEL, …
llama.cpp	LLAMA_CPP_MODEL — path to a local GGUF file
QTS	XLAI_QTS_MODEL_DIR, optional voice-clone inputs, vocoder tuning env vars
Hugging Face release	HF_TOKEN (for maintainers running the HF workflow)

Ignored OpenAI e2e tests load .env automatically when you run them locally.

Embeddings

• OpenAI-compatible one-shot embeddings default to OPENAI_EMBEDDING_MODEL when set, otherwise they fall back to OPENAI_MODEL.
• Gemini embeddings use GEMINI_EMBEDDING_MODEL when set; the default in .env.example is gemini-embedding-001.
• dimensions is supported for remote OpenAI-compatible and Gemini embeddings when the selected provider/model supports it.
• Local llama.cpp and browser transformers.js embeddings ignore provider-specific projection and return Unsupported if dimensions is requested.

Next steps

• Rust SDK — wire OpenAiConfig, LlamaCppConfig, or other backends in code
• Development / CI — how credentials are used in GitHub Actions