Configuration

The config section key is ai_llm. Environment prefix: LEX_AI_LLM__.

ClientConfig

Key	Type	Default	Env Var	Description
`enabled`	`bool`	`True`	`LEX_AI_LLM__ENABLED`	Enable the LLM subsystem
`provider`	`ModelProvider` (str enum)	`"openai"`	`LEX_AI_LLM__PROVIDER`	LLM provider name
`model`	`str`	`"gpt-4-turbo"`	`LEX_AI_LLM__MODEL`	Model name or identifier
`api_key`	`SecretStr \| None`	`None`	`LEX_AI_LLM__API_KEY`	API key for the provider
`api_base`	`str \| None`	`None`	`LEX_AI_LLM__API_BASE`	Custom API base URL
`temperature`	`float`	`0.7`	`LEX_AI_LLM__TEMPERATURE`	Sampling temperature (0.0–2.0)
`max_tokens`	`int \| None`	`None`	`LEX_AI_LLM__MAX_TOKENS`	Max tokens in response
`timeout`	`float`	`60.0`	`LEX_AI_LLM__TIMEOUT`	Request timeout in seconds
`enable_cache`	`bool`	`False`	`LEX_AI_LLM__ENABLE_CACHE`	Enable response caching
`cache_ttl`	`int`	`3600`	`LEX_AI_LLM__CACHE_TTL`	Cache TTL in seconds
`thinking`	`ThinkingConfig \| None`	`None`	`LEX_AI_LLM__THINKING__*`	Thinking/reasoning config
`extra`	`dict[str, Any]`	`{}`	`LEX_AI_LLM__EXTRA__*`	Provider-specific extra params

ThinkingConfig sub-fields

When thinking is set, the ThinkingConfig object supports:

Key	Type	Default	Description
`suppress`	`bool`	`False`	Actively suppress thinking tokens
`budget_tokens`	`int`	`10000`	Max tokens for reasoning
`level`	`str \| None`	`None`	Gemini 3 thinking level (`minimal`, `low`, `medium`, `high`)
`effort`	`str \| None`	`None`	OpenAI reasoning effort (`low`, `medium`, `high`)

Example YAML

ai_llm:
  provider: openai
  model: gpt-4o
  temperature: 0.3
  max_tokens: 2000
  timeout: 30
  enable_cache: true
  cache_ttl: 7200
  thinking:
    budget_tokens: 5000

Env-var overrides

export LEX_AI_LLM__PROVIDER=anthropic
export LEX_AI_LLM__MODEL=claude-sonnet-4-20250514
export LEX_AI_LLM__API_KEY=sk-ant-...
export LEX_AI_LLM__TEMPERATURE=0.0
export LEX_AI_LLM__ENABLE_CACHE=true

Provider extras

Install the provider-specific client library:

uv add "lexigram-ai-llm[openai]"    # openai, tiktoken
uv add "lexigram-ai-llm[anthropic]" # anthropic
uv add "lexigram-ai-llm[ollama]"    # ollama
uv add "lexigram-ai-llm[groq]"     # groq
uv add "lexigram-ai-llm[cohere]"   # cohere
uv add "lexigram-ai-llm[mistral]"  # mistralai