LLM providers

Built-in providers

Set one key in .env and Watchtower picks the provider in this order:

Key	Provider	Model variable	Default model
OPENROUTER_API_KEY	OpenRouter	OPENROUTER_MODEL_NAME	anthropic/claude-sonnet-5.5
OPENAI_API_KEY	OpenAI	OPENAI_MODEL_NAME	gpt-5.6
GEMINI_API_KEY	Google Gemini	GEMINI_MODEL_NAME	gemini-3.8-flash

Custom providers

Watchtower supports almost any provider through LangChain and LiteLLM. Override the defaults with --provider, --model, and --apikey:

python -m watchtower.main -t https://www.example.com \
  --provider=https://api.dgrid.ai/api/v1 \
  --model=anthropic/claude-opus-4.5 \
  --apikey "API_KEY"

--provider accepts:

• a URL, used as an OpenAI-compatible endpoint (any custom HTTP endpoint works this way);
• openai, anthropic, openrouter, or litellm;
• any other provider name that LangChain's init_chat_model understands, with LiteLLM as the fallback.

Important

--apikey takes the name of the variable that holds your key in .env (for example MY_GROQ_KEY), not the key itself. This keeps secrets out of your shell history.

The same settings can be put in .env:

WATCHTOWER_PROVIDER="https://api.custom.com/v1"
WATCHTOWER_MODEL="custom-model-v1"
WATCHTOWER_APIKEY_NAME="MY_CUSTOM_KEY"
MY_CUSTOM_KEY="sk-..."

Example output (using the httpx tool):

INFO: ==> Node Executed: [WORKER]
INFO: HTTP Request: POST https://api.dgrid.ai/api/v1/chat/completions "HTTP/1.1 200 OK"
INFO: ==> Node Executed: [ANALYST]
INFO:     - Updated state 'findings': [{'title': 'Overly Permissive CORS Configuration', 'severity': 'Medium', ...}, {'title': 'Missing Security Headers', 'severity': 'Low', ...}]

Supported providers include: Anthropic, OpenAI, OpenRouter, LiteLLM, Amazon Bedrock, Vercel AI Gateway, Moonshot AI, Mistral, MiniMax, OpenCode Zen, GLM models, Z.AI, Synthetic, and Qianfan, plus any OpenAI-compatible HTTP endpoint. See the full list at docs.litellm.ai/docs/providers.

Flagship models and reasoning

See Usage: flagship models and reasoning for --flagship, --effort, and --max-tokens.