conf.yaml

# Letta Configuration File
# Place at ~/.letta/conf.yaml, ./conf.yaml, or set LETTA_CONFIG_PATH
# Environment variables take precedence over config file values
#
# Top-level keys and their env var mappings:
#   letta:   -> LETTA_*
#   model:   -> Provider-prefixed (OPENAI_*, ANTHROPIC_*, etc.)
#   tool:    -> Prefix-based (E2B_*, MCP_*, TOOL_*, etc.)
#   datadog: -> DD_*

letta:
  # =============================================================================
  # Core Settings (LETTA_*)
  # =============================================================================
  debug: false
  # environment: ""

  # Default handles
  # default_llm_handle: ""
  # default_embedding_handle: ""

  # SSE Streaming
  enable_keepalive: true
  keepalive_interval: 50.0
  enable_cancellation_aware_streaming: true

  # =============================================================================
  # PostgreSQL (LETTA_PG_*)
  # =============================================================================
  pg:
    # db: ""
    # user: ""
    # password: ""
    # host: ""
    # port: ""
    # uri: ""
    pool_size: 25
    max_overflow: 10
    pool_timeout: 30
    pool_recycle: 1800
    echo: false

  # Connection pool settings (LETTA_POOL_*)
  pool:
    pre_ping: true
    use_lifo: true

  # Database settings (LETTA_DB_*)
  # db:
  #   max_concurrent_sessions: ""

  disable_sqlalchemy_pooling: true
  enable_db_pool_monitoring: true
  db_pool_monitoring_interval: 30

  # =============================================================================
  # Redis (LETTA_REDIS_*)
  # =============================================================================
  redis:
    # host: ""
    port: 6379

  # =============================================================================
  # Multi-Agent (LETTA_MULTI_AGENT_*)
  # =============================================================================
  multi_agent:
    send_message_max_retries: 3
    send_message_timeout: 1200
    concurrent_sends: 50

  # =============================================================================
  # OTEL / Observability (LETTA_OTEL_*, LETTA_CLICKHOUSE_*)
  # =============================================================================
  otel:
    # exporter_otlp_endpoint: ""
    preferred_temporality: 1

  clickhouse:
    # endpoint: ""
    database: otel
    username: default
    # password: ""

  disable_tracing: false
  llm_api_logging: true
  track_last_agent_run: false
  track_errored_messages: true
  track_stop_reason: true
  track_agent_run: true
  track_provider_trace: true

  # =============================================================================
  # Uvicorn (LETTA_UVICORN_*)
  # =============================================================================
  uvicorn:
    workers: 1
    reload: false
    timeout_keep_alive: 5

  # Runtime settings
  use_uvloop: false
  use_granian: false
  sqlalchemy_tracing: false
  event_loop_threadpool_max_workers: 43

  # =============================================================================
  # Experimental
  # =============================================================================
  use_vertex_structured_outputs_experimental: false
  use_asyncio_shield: true

  # =============================================================================
  # Lettuce (LETTA_USE_LETTUCE_*)
  # =============================================================================
  use_lettuce_for_file_uploads: false

  # =============================================================================
  # Batch Job Polling (LETTA_POLL_*, LETTA_BATCH_*)
  # =============================================================================
  enable_batch_job_polling: false
  poll_running_llm_batches_interval_seconds: 300
  poll_lock_retry_interval_seconds: 480
  batch_job_polling_lookback_weeks: 2
  # batch_job_polling_batch_size: ""

  # =============================================================================
  # LLM Timeouts (LETTA_LLM_*)
  # =============================================================================
  llm:
    request_timeout_seconds: 60.0
    stream_timeout_seconds: 600.0

  # =============================================================================
  # Pinecone (LETTA_PINECONE_*, LETTA_ENABLE_PINECONE, LETTA_UPSERT_PINECONE_INDICES)
  # =============================================================================
  enable_pinecone: false
  upsert_pinecone_indices: false
  pinecone:
    # api_key: ""
    source_index: sources
    agent_index: recall

  # =============================================================================
  # Turbopuffer (LETTA_TPUF_*, LETTA_USE_TPUF, LETTA_EMBED_*)
  # =============================================================================
  use_tpuf: false
  embed_all_messages: false
  embed_tools: false
  tpuf:
    # api_key: ""
    region: gcp-us-central1

  # =============================================================================
  # File Processing (LETTA_FILE_PROCESSING_*)
  # =============================================================================
  file_processing:
    timeout_minutes: 30
    timeout_error_message: "File processing timed out after {} minutes. Please try again."

  # =============================================================================
  # Letta Client (LETTA_DEFAULT_*)
  # =============================================================================
  default_base_url: http://localhost:8283
  # default_token: ""

  # =============================================================================
  # Agent Architecture
  # =============================================================================
  use_letta_v1_agent: false
  archival_memory_token_limit: 8192

  # =============================================================================
  # Security
  # =============================================================================
  no_default_actor: false
  # encryption_key: ""

  # =============================================================================
  # OCR
  # =============================================================================
  # mistral_api_key: ""

  # =============================================================================
  # Summarizer (LETTA_SUMMARIZER_*)
  # =============================================================================
  summarizer:
    mode: partial_evict_message_buffer_mode
    message_buffer_limit: 60
    message_buffer_min: 15
    enable_summarization: true
    max_summarization_retries: 3
    partial_evict_summarizer_percentage: 0.30
    evict_all_messages: false
    max_summarizer_retries: 3
    memory_warning_threshold: 0.75
    send_memory_warning_message: false
    desired_memory_token_pressure: 0.3
    keep_last_n_messages: 0

  # =============================================================================
  # Logging (LETTA_LOGGING_*)
  # =============================================================================
  logging:
    debug: false
    json_logging: false
    log_level: WARNING
    verbose_telemetry_logging: false

  # =============================================================================
  # Telemetry (LETTA_TELEMETRY_*)
  # =============================================================================
  telemetry:
    enable_datadog: false
    provider_trace_backend: postgres
    socket_path: /var/run/telemetry/telemetry.sock
    provider_trace_pg_metadata_only: false
    # source: ""

    # Datadog settings (LETTA_TELEMETRY_DATADOG_*)
    datadog:
      agent_host: localhost
      agent_port: 8126
      service_name: letta-server
      profiling_enabled: false
      profiling_memory_enabled: false
      profiling_heap_enabled: false
      # git_repository_url: ""
      # git_commit_sha: ""
      main_package: letta

# =============================================================================
# Model Settings (-> OPENAI_*, ANTHROPIC_*, AWS_*, etc.)
# =============================================================================
model:
  # Global settings
  global_max_context_window_limit: 128000
  inner_thoughts_kwarg: thinking
  default_prompt_formatter: chatml

  # OpenAI (-> OPENAI_*)
  openai:
    # api_key: ""
    api_base: https://api.openai.com/v1

  # Anthropic (-> ANTHROPIC_*)
  anthropic:
    # api_key: ""
    max_retries: 3
    sonnet_1m: false

  # Azure OpenAI (-> AZURE_*)
  azure:
    # api_key: ""
    # base_url: ""
    api_version: "2024-09-01-preview"

  # Google Gemini (-> GEMINI_*)
  gemini:
    # api_key: ""
    base_url: https://generativelanguage.googleapis.com/
    force_minimum_thinking_budget: false
    max_retries: 5
    timeout_seconds: 600.0

  # Google Vertex (-> GOOGLE_CLOUD_*)
  # google_cloud:
  #   project: ""
  #   location: ""

  # AWS Bedrock (-> AWS_*, BEDROCK_*)
  aws:
    # access_key_id: ""
    # secret_access_key: ""
    default_region: us-east-1

  bedrock:
    anthropic_version: bedrock-2023-05-31

  # OpenRouter (-> OPENROUTER_*)
  # openrouter:
  #   api_key: ""
  #   referer: ""
  #   title: ""
  #   handle_base: ""

  # Groq (-> GROQ_*)
  # groq:
  #   api_key: ""

  # Together (-> TOGETHER_*)
  # together:
  #   api_key: ""

  # DeepSeek (-> DEEPSEEK_*)
  # deepseek:
  #   api_key: ""

  # xAI/Grok (-> XAI_*)
  # xai:
  #   api_key: ""

  # Z.ai/ZhipuAI (-> ZAI_*)
  zai:
    # api_key: ""
    base_url: https://api.z.ai/api/paas/v4/

  # MiniMax (-> MINIMAX_*)
  # minimax:
  #   api_key: ""

  # Ollama (-> OLLAMA_*)
  # ollama:
  #   base_url: ""

  # vLLM (-> VLLM_*)
  # vllm:
  #   api_base: ""
  #   handle_base: ""

  # SGLang (-> SGLANG_*)
  # sglang:
  #   api_base: ""
  #   handle_base: ""

  # LM Studio (-> LMSTUDIO_*)
  # lmstudio:
  #   base_url: ""

  # OpenLLM (-> OPENLLM_*)
  # openllm:
  #   auth_type: ""
  #   api_key: ""

# =============================================================================
# Tool Settings (-> E2B_*, MCP_*, MODAL_*, TOOL_*, etc.)
# =============================================================================
tool:
  # E2B Sandbox (-> E2B_*)
  # e2b:
  #   api_key: ""
  #   sandbox_template_id: ""

  # Modal Sandbox (-> MODAL_*)
  # modal:
  #   token_id: ""
  #   token_secret: ""

  # Search Providers (-> TAVILY_*, EXA_*)
  # tavily:
  #   api_key: ""

  # exa:
  #   api_key: ""

  # Local Sandbox (-> TOOL_*)
  tool:
    # exec_dir: ""
    sandbox_timeout: 180
    # exec_venv_name: ""
    exec_autoreload_venv: true

  # MCP (-> MCP_*)
  mcp:
    connect_to_server_timeout: 30.0
    list_tools_timeout: 30.0
    execute_tool_timeout: 60.0
    read_from_config: false
    disable_stdio: true

# =============================================================================
# Datadog Agent Settings (-> DD_*)
# =============================================================================
# datadog:
#   site: ""
#   service: ""
#   version: ""
#
#   trace:
#     enabled: false
#     agent_url: ""
#     health_metrics_enabled: false
#
#   dogstatsd:
#     url: ""
#
#   logs:
#     injection: false
#
#   runtime:
#     metrics_enabled: false
#
#   appsec:
#     enabled: false
#     sca_enabled: false
#
#   iast:
#     enabled: false
#
#   exception:
#     replay_enabled: false
#
#   llmobs:
#     enabled: false
#     ml_app: ""
#
#   instrumentation:
#     install_type: ""
#
#   git:
#     repository_url: ""
#     commit_sha: ""
#
#   main_package: ""