Skip to content

LLM Configuration

Settings for controlling how LambChat interacts with language models.

Model Provider Keys

These are consumed by the underlying LLM SDK libraries directly (not by the Settings class):

VariableDescription
LLM_API_KEYDefault LLM API key (consumed by LiteLLM)
LLM_API_BASEDefault LLM API base URL (consumed by LiteLLM)
LLM_MODELDefault LLM model name, e.g. anthropic/claude-sonnet-4-6
ANTHROPIC_API_KEYAnthropic API key (consumed by langchain-anthropic)
ANTHROPIC_BASE_URLAnthropic-compatible API base URL

TIP

LambChat supports multi-model management through the UI. The env vars above set the default provider. Users can add additional providers and models at runtime through the settings panel.

Retry & Cache Settings

VariableDefaultDescription
DEFAULT_MODEL_ID(empty)Admin model configuration ID used as the default for new sessions and background jobs. Empty = first enabled model.
LLM_MAX_RETRIES3Retries after the initial call for timeout, network, rate-limit, and 5xx failures. 3 means up to 4 attempts.
LLM_RETRY_DELAY1.0Initial retry delay in seconds (exponential backoff).
LLM_REQUEST_TIMEOUT0Total seconds allowed for a complete non-streaming response. 0 or a negative value disables LambChat's non-streaming total timeout, so non-streaming waits indefinitely by default.
LLM_FIRST_EVENT_TIMEOUT30Seconds allowed for the first provider event of a streaming response. 0 or a negative value disables this deadline. After the first event, LambChat applies no total duration limit or chunk idle timeout.
LLM_MODEL_CACHE_SIZE50Model instance cache size. Prevents memory leaks from repeated instantiation.
LLM_REQUEST_HEADERS(empty)JSON object of request headers merged over the built-in anti-ban defaults (Claude Code style User-Agent/x-app for Anthropic protocol, opencode style User-Agent for OpenAI-compatible). Example: {"User-Agent": "my-agent/1.0"}. Per-model header overrides take precedence; Google protocol is not supported.
LLM_MAX_INPUT_TOKENS(none)Optional: context window size for DeepAgent auto-summarization.
LLM_TEMPERATURE(none)Optional: default temperature for LLM calls.
LLM_MAX_TOKENS(none)Optional: max output tokens for LLM calls.

DeepAgent Context Settings

VariableDefaultDescription
DEEPAGENT_DEFAULT_MAX_INPUT_TOKENS64000Default max input tokens for DeepAgent.

Example

bash
# .env
LLM_API_KEY=sk-your-api-key
LLM_API_BASE=https://api.openai.com/v1
LLM_MODEL=gpt-4o
LLM_MAX_RETRIES=3
LLM_RETRY_DELAY=1.0
LLM_REQUEST_TIMEOUT=0
LLM_FIRST_EVENT_TIMEOUT=30
LLM_MODEL_CACHE_SIZE=50