LLM configuration
Set SYSTEM_DEFAULT_LLM_MODEL (the model) and SYSTEM_DEFAULT_LLM_API_KEY (its API key) in backend/.env:
| Provider | Value |
|---|---|
| Google Gemini (default) | gemini-3.8-flash |
| OpenAI via LiteLLM | litellm:openai/gpt-6.1-sol |
| Anthropic via LiteLLM | litellm:anthropic/claude-sonnet-5-5 |
| Amazon Bedrock via LiteLLM | litellm:bedrock/global.anthropic.claude-sonnet-5-5 |
These are the system's default LLM. The backend writes them into System Settings every time it starts, so a variable that is set always wins over that screen; leave one unset to change that field on the screen instead — see LLM for the exact rules. A tenant can run on another model or a key of its own: see The tenant's LLM.
An agent uses these, in this order. Each field falls through on its own, so a tenant that sets only a model still uses the system's key:
| Order | Source |
|---|---|
| 1 | The tenant's own LLM, when a Super Admin has given the tenant one |
| 2 | The system's LLM on the System Settings screen |
| 3 | The built-in default model, gemini-3.8-flash |
With no key in either place, the provider's own variable — GOOGLE_API_KEY, OPENAI_API_KEY, ANTHROPIC_API_KEY or AWS_BEARER_TOKEN_BEDROCK — is read as before.
Every other backend setting has its own page under Configuration reference.
Gemini (default)
SYSTEM_DEFAULT_LLM_MODEL=gemini-3.8-flash
SYSTEM_DEFAULT_LLM_API_KEY=your_google_api_key_here
OpenAI (via LiteLLM)
SYSTEM_DEFAULT_LLM_MODEL=litellm:openai/gpt-6.1-sol
SYSTEM_DEFAULT_LLM_API_KEY=your_openai_api_key_here
Anthropic (via LiteLLM)
SYSTEM_DEFAULT_LLM_MODEL=litellm:anthropic/claude-sonnet-5-5
SYSTEM_DEFAULT_LLM_API_KEY=your_anthropic_api_key_here
Amazon Bedrock (via LiteLLM)
The boto3 and botocore[crt] packages this route needs ship with the backend, so there is no extra install step.
SYSTEM_DEFAULT_LLM_MODEL=litellm:bedrock/global.anthropic.claude-sonnet-5-5
SYSTEM_DEFAULT_LLM_API_KEY=your_bedrock_bearer_token_here