odysseus/src/settings.py

# src/settings.py
"""Centralized settings and features management.

Single source of truth for reading/writing data/settings.json and data/features.json.
All modules should import from here instead of accessing files directly.
"""

import json
import time
import logging
from typing import Any

from src.constants import SETTINGS_FILE, FEATURES_FILE

logger = logging.getLogger(__name__)

# Tiny TTL cache for settings/features. get_setting() is called on hot paths
# (every chat, every preprocess); without this it re-parses the JSON each call.
# Picks up edits within _CACHE_TTL seconds, which is fine for human-edited config.
_CACHE_TTL = 2.0
_settings_cache: tuple[float, dict] | None = None
_features_cache: tuple[float, dict] | None = None

def _invalidate_caches():
    global _settings_cache, _features_cache
    _settings_cache = None
    _features_cache = None

# ── Default values ──

DEFAULT_SETTINGS = {
    "image_gen_enabled": True,
    "image_model": "",
    "image_quality": "medium",
    "vision_model": "",
    "vision_enabled": True,
    # Ordered fallback chain for the Vision model (image analysis, OCR, tagging).
    "vision_model_fallbacks": [],
    # Public base URL used to build clickable deep-links in outgoing alerts
    # (e.g., urgency alert email). Example: "https://chat.example.com"
    "app_public_url": "",
    "tts_enabled": True,
    "tts_provider": "disabled",
    "tts_model": "tts-1",
    "tts_voice": "alloy",
    "tts_speed": "1",
    "stt_enabled": False,
    "stt_provider": "disabled",
    "stt_model": "base",
    "stt_language": "",
    "search_provider": "searxng",
    # Default fallback chain — when the primary provider fails or
    # rate-limits, we try DuckDuckGo next. Free, no API key required, so
    # safe to ship on by default for every user.
    "search_fallback_chain": ["duckduckgo"],
    "search_url": "",
    "search_result_count": 5,
    "brave_api_key": "",
    "google_pse_key": "",
    "google_pse_cx": "",
    "tavily_api_key": "",
    "serper_api_key": "",
    "research_endpoint_id": "",
    "research_model": "",
    "research_search_provider": "",
    "research_max_tokens": 16384,
    "agent_max_tool_calls": 0,
    "agent_input_token_budget": 6000,
    "agent_stream_timeout_seconds": 300,
    "task_endpoint_id": "",
    "task_model": "",
    "default_endpoint_id": "",
    "default_model": "",
    # Ordered fallback chain for the default chat model. Each entry is
    # {"endpoint_id": "...", "model": "..."}. If the primary model fails
    # before producing output (endpoint offline / errors), the chat
    # dispatch retries the next entry in order.
    "default_model_fallbacks": [],
    "utility_endpoint_id": "",
    "utility_model": "",
    # Ordered fallback chain for the Utility model (summarization, naming,
    # tidy actions, etc.).
    "utility_model_fallbacks": [],
    "teacher_model": "",
    "teacher_enabled": False,
    # Skills: minimum self-reported confidence for an auto-written (LLM-authored)
    # DRAFT skill to be injected into the agent prompt. Published skills always
    # qualify. Keeps low-confidence auto-skills out of context until they're
    # vetted/published. 0 disables the gate.
    "skill_autosave_min_confidence": 0.85,
    # Max relevant skills injected into the prompt for one request. The skills
    # library can grow beyond this; cleanup/retirement is an explicit review flow.
    "skill_max_injected": 3,
    # Reminders
    "reminder_channel": "browser",   # "browser" | "email" | "ntfy"
    "reminder_llm_synthesis": False,
    "reminder_ntfy_topic": "Reminders",
    "reminder_email_to": "",
    # Email triage scanner rules. Running/paused state and schedule live in
    # Tasks via the built-in `check_email_urgency` task.
    "urgent_email_prompt": (
        "Flag as urgent: explicit deadlines, time-sensitive requests, "
        "work-blocking issues, messages from people I report to, or anything "
        "where a delayed reply costs money/trust. Someone waiting outside, "
        "at the door, locked out, or unable to get in is urgent now. "
        "Newsletters, marketing, automated digests, and FYI-only updates are "
        "NOT urgent."
    ),
    # Keyboard shortcuts (action: key combination)
    "keybinds": {
        "search": "ctrl+k",
        "toggle_sidebar": "ctrl+b",
        "new_session": "ctrl+alt+n",
        "star_session": "ctrl+alt+s",
        "delete_session": "ctrl+alt+d",
        "admin_panel": "ctrl+shift+u",
        "cancel": "escape",
    },
}

DEFAULT_FEATURES = {
    "web_search": True,
    "deep_research": False,
    "memory": True,
    "document_editor": True,
    "rag": True,
    "sensitive_filter": True,
    "gallery": True,
}


# ── Settings (data/settings.json) ──

def load_settings() -> dict:
    """Load settings merged with defaults. Always returns a complete dict."""
    global _settings_cache
    now = time.monotonic()
    if _settings_cache and (now - _settings_cache[0]) < _CACHE_TTL:
        return _settings_cache[1]
    try:
        with open(SETTINGS_FILE, "r") as f:
            saved = json.load(f)
        merged = {**DEFAULT_SETTINGS, **saved}
    except (FileNotFoundError, json.JSONDecodeError):
        merged = dict(DEFAULT_SETTINGS)
    _settings_cache = (now, merged)
    return merged


def save_settings(settings: dict):
    """Persist settings to disk (atomic; see core.atomic_io)."""
    from core.atomic_io import atomic_write_json
    atomic_write_json(SETTINGS_FILE, settings, indent=2)
    _invalidate_caches()


def get_setting(key: str, default: Any = None) -> Any:
    """Read a single setting value."""
    return load_settings().get(key, default)


# Per-user settings (user prefs override the global admin default). Used for
# keys that a user is allowed to choose individually — currently the vision
# model + image-generation model. The owner argument is the authed username
# resolved by FastAPI deps; an empty/None owner falls through to the global.
_PER_USER_KEYS = {
    "vision_model", "vision_enabled", "vision_model_fallbacks",
    "image_model", "image_gen_enabled", "image_quality",
    # Default chat endpoint / model — without per-user resolution every new
    # account inherited whatever the most-recent admin picked, which then
    # got injected into the chat composer on first open.
    "default_endpoint_id", "default_model", "default_model_fallbacks",
    "utility_endpoint_id", "utility_model", "utility_model_fallbacks",
    "research_endpoint_id", "research_model",
}


def get_user_setting(key: str, owner: str = "", default: Any = None) -> Any:
    """Resolve `key` from the caller's per-user prefs first, falling back to
    the global setting. Only the small whitelist in `_PER_USER_KEYS` is
    eligible — for any other key this is equivalent to `get_setting(key)`.

    Falls back gracefully if the prefs module can't be imported (cycle/early
    boot) — admin-global settings keep working.
    """
    if owner and key in _PER_USER_KEYS:
        try:
            from routes.prefs_routes import _load_for_user
            prefs = _load_for_user(owner) or {}
            if key in prefs and prefs[key] not in (None, ""):
                return prefs[key]
        except Exception:
            pass
    return get_setting(key, default)


# ── Features (data/features.json) ──

def load_features() -> dict:
    """Load feature flags merged with defaults."""
    global _features_cache
    now = time.monotonic()
    if _features_cache and (now - _features_cache[0]) < _CACHE_TTL:
        return _features_cache[1]
    try:
        with open(FEATURES_FILE, "r") as f:
            saved = json.load(f)
        merged = {**DEFAULT_FEATURES, **saved}
    except (FileNotFoundError, json.JSONDecodeError):
        merged = dict(DEFAULT_FEATURES)
    _features_cache = (now, merged)
    return merged


def save_features(features: dict):
    """Persist feature flags to disk (atomic)."""
    from core.atomic_io import atomic_write_json
    atomic_write_json(FEATURES_FILE, features, indent=2)
    _invalidate_caches()