# NoteWise docs for LLMs

NoteWise is a Python CLI (`>=3.10`) that converts YouTube videos, playlists, and batch files into local study notes through LiteLLM providers. Current documented version: `1.4.0`.

Canonical public docs base: `https://notewise.click/docs`

Canonical routes:

- Overview: `/docs`
- Install: `/docs/start/install`
- Quickstart: `/docs/start/quickstart`
- Process videos: `/docs/use/process`
- Playlists and batches: `/docs/use/playlists-batches`
- Private videos and Docker: `/docs/use/private-docker`
- Configuration: `/docs/config/configuration`
- Models and providers: `/docs/config/providers`
- OAuth login: `/docs/config/oauth`
- CLI commands: `/docs/operate/commands`
- Cache, logs, history, stats: `/docs/operate/cache-logs-history`
- Troubleshooting: `/docs/operate/troubleshooting`
- Pipeline and output: `/docs/understand/pipeline-output`
- Storage, events, errors: `/docs/understand/storage-events`
- Website and docs: `/docs/understand/website-docs`
- Development: `/docs/understand/development`

Source truth:

- Default model: `gemini/gemini-2.5-flash`
- Default output directory: `./output`
- Default output format: `md`; supported formats: `md`, `html`, `pdf`, `docx`
- Temperature: `0.7`
- Max concurrent videos: `5`
- YouTube requests per minute: `10`
- Chunk size / overlap: `4000` / `200`
- Max concurrent chapters: `3`

CLI commands:

- `process`, `setup`, `config`, `config-path`, `version`, `update`, `stats`, `history`, `info`, `doctor`, `edit-config`, `auth login`, `cache`, `logs`

Important `process` flags:

- `--model/-m`, `--output/-o`, `--format`, `--language/-l`, `--target-language`, `--temperature/-t`, `--max-tokens/-k`, `--throttle`, `--force/-F`, `--no-ui`, `--verbose/-v`, `--quiz`, `--use-combine-chunk`, `--export-transcript`, `--timestamps`, `--chapter-directory-output`, `--cookie-file/--cookies`

Config truth:

- Config file: `~/.notewise/config.env`
- `NOTEWISE_HOME` changes the state root.
- Practical precedence: CLI flags, then environment variables, then config file, then defaults.
- Common keys: `DEFAULT_MODEL`, `OUTPUT_DIR`, `MAX_CONCURRENT_VIDEOS`, `YOUTUBE_REQUESTS_PER_MINUTE`, `TEMPERATURE`, `MAX_TOKENS`, `YOUTUBE_COOKIE_FILE`, provider API/auth keys.
- `chunk_size`, `chunk_overlap`, and `max_concurrent_chapters` are code-default settings, not normal config-file keys.

Provider truth:

- Gemini `GEMINI_API_KEY`
- OpenAI `OPENAI_API_KEY`
- Anthropic `ANTHROPIC_API_KEY`
- Groq `GROQ_API_KEY`
- xAI `XAI_API_KEY`
- Mistral `MISTRAL_API_KEY`
- Cohere `COHERE_API_KEY`
- DeepSeek `DEEPSEEK_API_KEY`
- OpenRouter `OPENROUTER_API_KEY`
- Azure `AZURE_API_KEY` or `AZURE_OPENAI_API_KEY`
- Vercel AI Gateway `VERCEL_AI_GATEWAY_API_KEY`
- Together `TOGETHERAI_API_KEY`
- Fireworks `FIREWORKS_AI_API_KEY`
- Perplexity `PERPLEXITYAI_API_KEY`
- Cloudflare needs `CLOUDFLARE_API_KEY` and `CLOUDFLARE_ACCOUNT_ID`
- Databricks needs `DATABRICKS_API_KEY` and `DATABRICKS_API_BASE`
- Bedrock uses the AWS credential chain

OAuth truth:

- ChatGPT login: `notewise auth login chatgpt`, safe model `chatgpt/gpt-5.2`
- GitHub Copilot login: `notewise auth login github_copilot`, safe model `github_copilot/gpt-5-mini`
- `codex` is a CLI alias for ChatGPT login.
- Token dirs default under `<state>/oauth/...` unless `CHATGPT_TOKEN_DIR` or `GITHUB_COPILOT_TOKEN_DIR` is set.

Output truth:

- Cache check happens before metadata/transcript unless `--force` is used.
- Videos with YouTube chapters use chapter-aware generation.
- Chapter videos default to bundled notes; per-chapter folders require `--chapter-directory-output`.
- PDF uses fpdf and Latin-compatible text; unsupported Unicode falls back to Markdown.
- `cache clear` deletes the SQLite DB and sidecars, not generated notes.
- Logs are structlog key-value text, not JSON.

Security guidance: never request or expose API keys, OAuth tokens, cookies, raw LLM prompts, or provider payloads.