← Back to Index

Configuration Guide

Overview

YAMLLM requires a YAML configuration file with specific sections to define LLM settings and behaviors. This guide details all required configuration sections and their options.

Required Configuration Sections

A complete YAMLLM configuration must include these sections:

# LLM Provider Settings
# Note: Do not include API keys in this file. Pass them via environment
# variables to your application and into the constructor.
provider:
  name: "mistralai"  # Required: openai, google, deepseek, or mistralai
  model: "mistral-small-latest"  # Required: model identifier
  base_url: "https://api.mistral.ai/v1/"  # Optional: custom endpoint

# Model Configuration
model_settings:
  temperature: 0.7
  max_tokens: 1000
  top_p: 1.0
  frequency_penalty: 0.0
  presence_penalty: 0.0
  stop_sequences: []
  
# Request Settings
request:
  timeout: 30
  retry:
    max_attempts: 3
    initial_delay: 1
    backoff_factor: 2
    
# Context Management
context:
  system_prompt: "You are a helpful assistant"
  max_context_length: 16000
  memory:
    enabled: true
    max_messages: 10
    session_id: 'MySession'
    conversation_db: "memory/conversation_history.db"
    vector_store:
      index_path: "memory/vector_store/faiss_index.idx"
      metadata_path: "memory/vector_store/metadata.pkl"
      top_k: 2
    
# Output Formatting
output:
  format: "text"  # text, json, or markdown
  stream: true

# Logging Configuration
logging:
  level: "INFO"
  file: "yamllm.log"
  format: "%(asctime)s - %(name)s - %(levelname)s - %(message)s"

# Tool Management
tools:
  enabled: false
  tool_timeout: 10
  tool_list: ['calculator', 'web_search', 'weather']

# Safety Settings
safety:
  content_filtering: true
  max_requests_per_minute: 60
  sensitive_keywords: []

Section Details

Provider Settings

Setting Type Required Description
name string Yes LLM provider name
model string Yes Model identifier
api_key string Yes API key (use env vars)
base_url string No Custom API endpoint

Model Settings

Setting Type Required Default Description
temperature float Yes 0.7 Response randomness
max_tokens integer Yes 1000 Max response length
top_p float Yes 1.0 Nucleus sampling
frequency_penalty float Yes 0.0 Token frequency penalty
presence_penalty float Yes 0.0 Token presence penalty
stop_sequences list Yes [] Stop sequence tokens

Request Settings

Setting Type Required Default Description
timeout integer Yes 30 Request timeout (seconds)
retry.max_attempts integer Yes 3 Max retry attempts
retry.initial_delay integer Yes 1 Initial retry delay
retry.backoff_factor integer Yes 2 Retry backoff multiplier

Context Management

Setting Type Required Default Description
system_prompt string Yes - System context
max_context_length integer Yes 16000 Max context tokens
memory.enabled boolean Yes true Enable memory
memory.max_messages integer Yes 10 History size
memory.conversation_db string Yes - SQLite DB path
memory.vector_store.* object Yes - Vector store settings

Output Settings

Setting Type Required Default Description
format string Yes “text” Response format
stream boolean Yes true Enable streaming

Logging Settings

Setting Type Required Default Description
level string Yes “INFO” Log level
file string Yes “yamllm.log” Log file path
format string Yes - Log format string
console boolean No false Also log to console
rotate boolean No false Enable rotating file handler
rotate_max_bytes integer No 1048576 Rotate when file exceeds this size
rotate_backup_count integer No 3 Number of rotated files to keep

Example:

logging:
  level: "INFO"
  file: "yamllm.log"
  format: "%(asctime)s - %(name)s - %(levelname)s - %(message)s"
  console: true
  rotate: true
  rotate_max_bytes: 2097152
  rotate_backup_count: 5

Tool Settings

Setting Type Required Default Description
enabled boolean Yes false Enable tools
tool_timeout integer Yes 10 Tool timeout
tool_list list Yes [] Available tools
packs list No [] Tool packs shortcut (e.g., [“common”, “web”])

Embeddings Settings

Optional section to control embeddings provider/model. If omitted, the selected provider is used when it supports embeddings; otherwise, OpenAI is used as a lazy fallback.

Setting Type Required Default Description
provider string or null No null Embeddings provider name (e.g., “openai”)
model string No “text-embedding-3-small” Embeddings model name

Example:

embeddings:
  provider: null
  model: "text-embedding-3-small"

Safety Settings

Setting Type Required Default Description
content_filtering boolean Yes true Enable filtering
max_requests_per_minute integer Yes 60 Rate limit
sensitive_keywords list Yes [] Blocked keywords

Environment Variables

Use environment variables for sensitive data:

provider:
  api_key: ${PROVIDER_API_KEY}

Provider-Specific Examples

See the .config_examples directory for complete provider-specific configurations:

Use packs to enable curated sets of tools without listing each by name. You can combine packs and explicit tool_list.

Example:

tools:
  enabled: true
  tool_timeout: 10
  packs: ["common", "web"]
  tool_list: ["unit_converter"]  # extra overrides