Skip to content

system.md

Latest commit

 

History

History
121 lines (91 loc) · 4.32 KB

system.md

File metadata and controls

121 lines (91 loc) · 4.32 KB
Raw

System Settings Guide

system.yaml is loaded from the project's .agentforge/settings/system.yaml and exposed via Config().data['settings']['system'].

Location

<project_root>/.agentforge/settings/system.yaml

Schema Overview

persona:
  enabled: true       # Load persona files from .agentforge/personas/
  name: default_assistant # Default persona filename (without .yaml)
  static_char_cap: 8000  # Max character length for persona markdown (0 disables truncation)

debug:
  mode: false         # If true, uses simulated_response instead of real LLM calls
  save_memory: false  # In debug mode, whether to save cog memory to storage
  simulated_response: "Text designed to simulate an LLM response for debugging purposes"

logging:
  enabled: true                 # Toggle all logging on or off
  console_level: warning        # Minimum severity for console output
  folder: ./logs                # Relative folder for log files
  create_missing_files: true    # Create runtime files for unlisted logger categories
  files:                        # Per-logger file-level overrides
    agentforge: debug
    model_io: debug

misc:
  on_the_fly: true   # Reload YAML configs at runtime for dynamic updates

audio:
  autoplay: true      # Play generated audio files after saving
  save_files: false   # Persist audio files or use a temporary directory
  save_dir: ./audio_files

paths:
  files: ./files     # Read/write directory available to agents

persona

  • enabled (bool): Toggle persona loading. Default true.
  • name (string): Persona filename (no .yaml). Default default_assistant.
  • static_char_cap (int): Maximum character length for persona markdown loaded from .agentforge/personas/. If set to 0, truncation is disabled. Default: 8000.
  • Behavior: When enabled, Config loads .agentforge/personas/<name>.yaml. Agents can override via their own persona key.

debug

  • mode (bool): Enable debug mode to bypass real LLM calls.
  • save_memory (bool): If debug mode is on, decide whether cogs should persist memory during tests.
  • simulated_response (string): Text returned in place of real model output.

logging

  • enabled (bool): Globally enable or disable logging.
  • console_level (string): One of critical, error, warning, info, debug.
  • folder (string): Path for writing log files, relative to project root.
  • create_missing_files (bool): When true, logger categories requested by code but missing from files get dedicated runtime files at warning level without editing system.yaml. When false, those requests route through the configured fallback logger.
  • files (map[string,string]): Keys are logger names; values are minimum log level for that file.

misc

  • on_the_fly (bool): When true, AgentForge re-reads YAML files before each run for rapid iteration.

audio

  • autoplay (bool): Play generated audio files after they are saved.
  • save_files (bool): Persist generated audio files when true; use temporary files when false.
  • save_dir (string): Directory for saved audio files when persistence is enabled.

paths

  • files (string): Default directory for agent I/O operations. You can add extra entries (e.g., paths.temp) and they will appear under settings.system.paths.

Accessing System Settings

from agentforge.config import Config
system_settings = Config().data['settings']['system']

Examples

# Run in debug with a custom simulated response
debug:
  mode: true
  simulated_response: "I'm a simulated response from debug mode."

# Disable real LLM calls but still save memory
debug:
  mode: true
  save_memory: true

# Change global console log level but keep file logging enabled
logging:
  console_level: info

# Route unlisted logger categories through agentforge instead of creating new files
logging:
  create_missing_files: false

# Add a custom paths entry for user content
paths:
  files: ./files
  user_content: ./user_content

Related Guides

Need Help?

If you have questions or need assistance, feel free to reach out: