""" App-wide configuration. Everything is either an environment variable (set once, at deploy time) or a row in the `settings` table (editable at runtime via the setup wizard / settings page) - never a hardcoded path or host. """ import os from pathlib import Path # Deploy-time config (env vars) - where things live on disk inside the container. DATA_DIR = Path(os.environ.get("APP_DATA_DIR", "/data")) UPLOAD_DIR = DATA_DIR / "audio" AUDIO_DIR = DATA_DIR / "audio" TRANSCRIPT_DIR = DATA_DIR / "transcriptions" NOTES_DIR = DATA_DIR / "notes" DB_PATH = DATA_DIR / "app.db" CAMPAIGNS_DIR = DATA_DIR / "campaigns" def campaign_dir(campaign_id: str) -> Path: return CAMPAIGNS_DIR / campaign_id def campaign_audio_dir(campaign_id: str) -> Path: return campaign_dir(campaign_id) / "audio" def campaign_transcript_dir(campaign_id: str) -> Path: return campaign_dir(campaign_id) / "transcriptions" def campaign_notes_dir(campaign_id: str) -> Path: return campaign_dir(campaign_id) / "notes" for d in (UPLOAD_DIR, AUDIO_DIR, TRANSCRIPT_DIR, NOTES_DIR): d.mkdir(parents=True, exist_ok=True) # Runtime-editable settings (stored in DB, these are just first-run defaults). # NOTE: this flat key/value schema is the single source of truth for settings - # it's what every consumer (pipeline/summarize.py, pipeline/transcribe.py, # routers/models.py, routers/diagnostics.py) actually reads. The settings # router and frontend must mirror these exact keys - a nested schema would # silently disconnect the Settings/Setup UI from the pipeline. DEFAULT_SETTINGS = { # Onboarding "onboarding_completed": "false", # Transcription "whisper_model": "medium", # tiny|base|small|medium|large-v3 - user picks based on their hardware "whisper_compute_type": "int8", "hf_token": "", # required for diarization (pyannote gated models) # Summarization backend: "ollama" (local) or "api" (hosted, OpenAI-compatible) "llm_mode": "ollama", "ollama_host": "http://localhost:11434", "ollama_model": "qwen2.5:7b", "api_base_url": "https://api.openai.com/v1", "api_key": "", "api_model": "gpt-4o-mini", # Chunking for long transcripts (map-reduce summarization) "chunk_word_target": "2500", # Optional world context injected into every summarization prompt "world_context": "", "world_context_path": "", # Player recap style: "story" | "diary" | "bullets" | "custom" "player_recap_style": "story", "player_recap_custom_prompt": "", }