The settings window is Findex's control room
Open it with the gear button in the search window or Settings in the tray menu. Its six tabs cover everything from window mode to the OCR languages used for image-only PDFs. Saved changes apply immediately.
Tab 1
General — language, shortcut, window behavior, extensions
| Setting | What it does |
|---|---|
| Language | UI, notification, and Luna response language. 12 languages. |
| Global shortcut | The summon key (default Ctrl+Shift+F). Click the field and press your combination. |
| Engine port | Local port for the internal engine (default 8765). Change only on a port conflict. |
| Start at login | Starts quietly in the tray after boot. |
| Window mode | Simple mode is a compact search bar that stays off the taskbar and hides when it loses focus. Normal mode is the full Command Center with filters, recent documents, preview, taskbar presence, and minimize/maximize controls. You can switch modes from the search header too. |
| Hide after opening a result | Dismisses the window when you open a file; off keeps it visible. |
| MCP integration | Enables external MCP tools and the Connection Center app. |
| Notifications | Master switch for app notifications, including to-do and schedule reminders. |
| Theme | Light / Dark / Glass / Dark Glass. |
Tab 2
Folders — what gets indexed, and progress
- Add/remove folders: adding starts a scan automatically; removing cleans up that folder's index.
- Folder profile: each folder shows its drive type (fixed/removable/network) and scan tier so you can spot slow sources.
- Progress: per-folder scan and indexing progress, plus a button to open the preparation window.
Tab 3
Analysis — what to read and how
- Precise-search file types: only checked formats are content-indexed. Unchecking unused formats lightens indexing.
- Document-type rules: add or edit label + keywords rules (e.g. “Report = report, results, summary”). The document-type dropdown in search follows these rules.
- Max file size: documents above the limit are skipped for content analysis. Lower it if giant PDFs slow indexing.
Tab 4
LLM — managing the AI connection
- LLM on/off: the switch at the top right toggles all LLM features. Off disables Luna; search keeps working.
- Provider: SAMI-GPT, OpenAI, Claude, Gemini, OpenRouter, Ollama, LM Studio, or OpenAI-compatible. Each provider's settings are stored separately, so switching back never loses your keys.
- Auth: API key by default; OpenAI also supports ChatGPT OAuth sign-in.
- Fetch models: pulls the available model list from the provider.
- Connection test: runs a real streaming request to validate key, URL, and model — the first thing to try when Luna is not working.
Tab 5
Performance — resource usage that fits your PC
Based on your detected memory and CPU, four presets are available. Recommended is right for most people; use Custom only when you want to hand-tune the limits.
| Preset | When to use |
|---|---|
| Eco | Minimal impact on other work — for modest PCs or always-on use. |
| Recommended | The default — auto-tuned to your system. |
| Max | High limits for fast indexing — good for the initial pass over large libraries. |
| Custom | Hand-tune memory cap, CPU limit, concurrency, per-document timeout, Excel rows per sheet, and more. |
Idle Boost
To speed up indexing only while you are away, enable Idle Boost: limits are relaxed only after N idle minutes and within the allowed time window (e.g. 19:00–08:00). Any mouse or keyboard activity restores normal limits immediately.
Tab 6
Engine — diagnostics, restart, reset, OCR
| Action | What it does / caution |
|---|---|
| Start / Restart engine | Starts or restarts the backend runtime. A restart briefly interrupts running search and indexing. |
| File scan | Rescans all watched folders — useful when network-folder changes were missed. |
| Engine test | Runs a self-diagnostic across health, search, and document search. |
| LLM stream test | Fetches a real response stream with the current LLM settings. |
| Clear chat history | Deletes Luna sessions only; index, folders, and LLM settings stay. |
| Reset search data | Wipes metadata and indexes, then rescans and re-indexes automatically. Results may look empty until the rescan finishes. |
| Reset everything | Erases sessions, indexes, folders, LLM connection, and settings, returning to the setup wizard. Irreversible — use with care. |
The status area also shows retry countdowns and workload details (e.g. queued PDF parses). For recurring errors, see the log guide in the Troubleshooting chapter.
Engine tab · OCR
Search image-only PDFs with official PP-OCRv5 recognition models
A scanned PDF without a text layer needs OCR before its contents can be indexed. OCR is off by default. In the Engine tab, enable OCR, select recognition languages, and save. Findex runs recognition locally and loads only the smallest model set required by your selection.
Bundled models and languages
| Model | Recognition coverage |
|---|---|
| default · PP-OCRv5 | Simplified and Traditional Chinese, English, and Japanese |
| ko · Korean PP-OCRv5 | Korean, English, and digits |
- Load only what you select: Korean-only work can use the ko model alone; selecting Chinese or Japanese adds the default model. Findex keeps the loaded set minimal.
- Run mode: Idle-only processes OCR while the PC is unattended; Balanced continues during active use while limiting resource pressure.
- Luna priority: OCR requested for Luna's current answer may be processed ahead of normal queued work.
Add a custom language model
Create a lowercase language-code folder under %LOCALAPPDATA%\Findex\ocr-models, then place the two files from an official PaddleOCR ONNX recognition model together in that folder. The shared detector, det.onnx, belongs directly under ocr-models.
ocr-models\det.onnxocr-models\default\inference.onnx + inference.ymlocr-models\ko\inference.onnx + inference.ymlocr-models\<lang>\inference.onnx + inference.yml
- Add the two-file pairBoth
inference.onnxandinference.ymlare required. Incomplete folders are ignored. - Refresh OCR modelsFindex discovers the new language folder and adds it to the choices. Discovery alone does not enable it.
- Select the language and saveOnly checked languages are loaded and used for OCR.