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: ""