Skip to Content
cuelo.

07Reference

Environment variables

Credentials and every CUELO_* setting.

Cuelo loads .env from the current directory (or cuelo-serve --env-file). Real environment variables take precedence.

Provider credentials

VariableUsed by
SMALLEST_API_KEYSmallest STT and TTS
DEEPGRAM_API_KEYDeepgram STT and TTS
ELEVENLABS_API_KEYElevenLabs STT and TTS
CARTESIA_API_KEYCartesia STT and TTS
INWORLD_API_KEYInworld STT and TTS
ASSEMBLYAI_API_KEYAssemblyAI STT
SONIOX_API_KEYSoniox STT
ANTHROPIC_API_KEYAnthropic
OPENAI_API_KEYOpenAI, and openai-compatible by default
CEREBRAS_API_KEYCerebras
OPENROUTER_API_KEYOpenRouter

Any provider can read a different variable with api_key_env in config.cuelo, for example two Anthropic keys for the talker and the thinker.

Server

VariableDefault
CUELO_HOST127.0.0.1cuelo-serve listen address
CUELO_PORT4141cuelo-serve port
CUELO_WORKERS1cuelo-serve processes
CUELO_API_KEY—default agent: required API key(s), comma-separated
CUELO_CORS—default agent: allowed browser origins, comma-separated

Projects set keys and CORS in config.cuelo (serve.api_keys, serve.cors) instead.

Default endpoint

CUELO_STT, CUELO_TTS, CUELO_VOICE, CUELO_LLM, CUELO_MODEL, CUELO_TURNS, CUELO_BARGE_IN, CUELO_AGENT_BC, CUELO_BC_HOLD and CUELO_DB pick the default agent's stack. See Default endpoint.

Models and performance

VariableDefault
CUELO_MODELS~/.cache/cuelo/modelswhere turn-taking weights are cached
CUELO_TORCH_THREADS1PyTorch threads per process (MaAI). More threads help one call, not many.

Development

Variable
CUELO_ENGINE_PATHgenerated projects use this local engine checkout (editable), like --engine-path
CUELO_ENGINE_GIToverride the engine's git source in generated projects
CUELO_ENGINE_URLthe engine the dashboard dev server proxies to (default http://localhost:4141)

On this page