# How to use Qube Practical workflows for voice, Library (RAG), and Memory Manager. For control-by-control reference, use in-app **Library → Qube** or **`@[tool:help]`**. --- ## Voice interaction 1. **Set up inference** — In **Settings → AI & Models**, choose **Internal Engine** (after loading a `.gguf` in **Model Manager**) or **External Server** and start LM Studio / Ollama if using external mode. 2. **Choose audio devices** — **Settings → Voice & Audio** → **Audio Input** / **Audio Output**. 3. **Enable listening** — Turn on **Enable Voice Input** in the Conversations tools panel, or use **push-to-talk** beside the composer. 4. **Wake word (optional)** — **Settings → Voice & Audio → Wakeword** → pick **Active Wakeword**, download models if prompted, test in **Open Wakeword Test Lab**. 5. **Speak** — Qube uses voice-activity detection: it listens while you talk and processes after a silence window (configurable in Settings). **Barge-in:** you can interrupt TTS mid-sentence by speaking again (when voice input is active). --- ## Library (document retrieval) Ask questions grounded in your own files. 1. Open **Library** and import documents (**PDF**, **EPUB**, **TXT**, **MD**). Qube parses and embeds them locally. 2. Ground a turn: - Attach **`@[tool:library]`** in the composer, and/or - Enable **Local Knowledge Base** in the tools panel, and/or - Use natural phrasing — the cognitive router generalizes from a few trigger examples in **Settings → Knowledge**. 3. Ask your question. Replies include numbered citations (e.g. `[1]`) linking to source chunks. 4. **Follow-ups** — conversation context carries forward; you often do not need to re-attach Library for the next question on the same topic. **Pro (optional):** With a **Qube Pro** license, each **Library (+)** import lets you choose **Normal** or **Precision indexing** (gem badge on precision-indexed docs). **Settings → Knowledge → Library Pro depth** offers **Default precision ingest on import** and **Precision retrieval** (rerank on each query). See [library-pro-depth.md](library-pro-depth.md) or in-app **Library → Qube → faq/library-pro-depth.md**. | Dark | Light | | :---: | :---: | | ![Library dark](../../assets/screenshots/qube_library_dark_mode.png) | ![Library light](../../assets/screenshots/qube_library_light_mode.png) | --- ## Live sources and research Attach **`@` tools** in the composer for capabilities beyond your Library: | Tool | Use for | |------|---------| | `@trusted` | General facts from allowlisted sources | | `@evidence` | Scientific literature | | `@finance` | SEC filings | | `@legal` | U.S. case law | | `@research` | Multi-step async evidence report | | `@internet` | Live web search | See **Library → Qube → reference/composer-tools.md** for the full list. --- ## Memory Manager Qube learns preferences and facts over time in the background. You do not have to open Memory Manager — decay and reflection help keep the store healthy — but it gives full editorial control. 1. Open **Memory Manager** in the main navigation. 2. Review top sections: **Promotion candidates**, **Almost promoted**, **Flagged for review** (when applicable). 3. Filter by **tier**, **category**, or search text; toggle **Flagged only**. 4. **Edit**, **Flag**, or **Delete** individual rows. **Delete** also adds the fact to a negative list so it is not re-extracted from similar chat. 5. **Export visible** writes Markdown to `~/.qube/exports/`. Optional background workers: **Settings → Memory** (promotion, consolidation). --- ## Desktop Companion Optional floating orb for quick voice turns without raising the main window. **Settings → Desktop Companion** — enable the companion, set position, fullscreen suppression, and optional commentary. --- ## Getting help - **`@[tool:help]`** in chat - **Library → Qube** — full corpus index at `00-index.md` - **`?` buttons** on each screen — guided tours --- ## Related - [Install from source](install-from-source.md) - [System requirements](system-requirements.md) - [Cognitive router (technical)](../cognitive_router.md)