core.modes

Mode modules: toggleable mode registry (single source of truth) with callback propagation and persistent settings.

Mode Registry

core.modes.mode_registry

Mode registry and command parsing.

Provides ModeRegistry as the single source of truth for all ten mode flags. The previous pattern (updating globals + config module separately) has been replaced by a callback-based sync mechanism via set_mode_flag().

process_mode_commands parses voice commands and delegates flag toggles to the registry.

ModeRegistry

Central store for all mode flags.

Callers read via .get(name) and write via .set(name, value). Changes propagate to main.py globals and config module via callbacks (legacy compat path).

ModeFlag dataclass

Backing store for one mode toggle with optional callbacks.

process_mode_commands(user_text: str, agents: list, registry: ModeRegistry, language_model: str, analysis_model: str, logger, torch, set_flag: Callable[[str, bool], None] | None = None) -> None

Parse spoken mode-toggle commands.

Dependencies are passed explicitly so the function is testable.

When set_flag is provided it's used instead of registry.set() so the caller can keep globals and config-module attributes in sync (three-way sync). Without it only the registry is updated.

Settings Persistence

core.modes.settings_persistence

Settings persistence layer for cross-restart durability.

Wires SQLite settings storage into the control update flow and provides startup restore. All UI-driven configuration changes (sampling params, mode toggles, model selections, agent state) are persisted to SQLite and restored on next launch. System prompts (general and per-agent) are persisted during a session but reset to config defaults on restart so users don't have to manually clear them after experiments.

Key naming scheme

sampling.{key} - temperature, top_p, top_k, etc. mode.{flag_name} - cloud_mode, mute_mode, analysis_mode, etc. agent.{name}.{setting} - active (restored), system_prompt1 (persisted only) model.{role} - language_model, analysis_model, etc. prompt.{key} - system_prompt_chat_single_agent, etc. (persisted only)

persist_control_update(update: dict[str, Any], agents: list[Any], store) -> None async

Persist changed settings to SQLite after _apply_control_update applies them.

Each call is wrapped in try/except so persistence failures never break runtime behavior.

restore_settings_to_runtime(store, config_module: Any, agents: list[Any], globals_dict: dict[str, Any] | None = None) -> None async

Restore persisted settings from SQLite to runtime config/globals/agents.

Parameters

store : SQLiteStore (or any object with list_settings() async method). config_module : The config.config module object – attributes are set via setattr. agents : list[Agent] Built agent instances for agent-specific setting restore. globals_dict : dict | None Optional: the module-level globals() dict from main.py. If provided, mode flags and model roles will be restored here too (maintains three-way sync invariant).