在 Pro 面板的 AI 设置中配置你自己的 OpenAI 兼容接口（baseURL + apiKey + model，支持本地 Ollama）后，即可在 AI 分类下自然语言提问并获得回答。请求直连你配置的接口，不经过 BestNav 官方服务器。
Aggregated Search
The top search bar (Ctrl/Cmd+K) covers 12 categories: Web, Music, Movies, Books, Docs, Papers, Images, Video, Files, Software, Skills and AI. Whenever a category has at least two executable sources, an Aggregate button appears and runs them concurrently, rendering results as they land. In Console and Desktop, Ctrl/Cmd+K opens the full-page search view.
Yes — if none is set the source simply does not join, quick-jump links still work
Music
iTunes, Netease (plus your own Navidrome library)
No
Movies
TMDB, TMDB Chinese (plus your Jellyfin / Emby library)
TMDB needs a key; it drops out when unset
Books
Google Books, Open Library
No
Docs
Wikipedia
No
Papers
arXiv, Semantic Scholar
No
Images
Wikimedia, Pixabay (plus your own Immich library)
Pixabay needs a key
Video
bilibili, Wikimedia
No
Software
GitHub + Docker Hub + Gitee (three run in parallel internally)
No
Skills
skills.sh + Smithery + GitHub (three run in parallel internally)
No
At most two sources per category join an aggregate, to keep the first screen fast — total latency is roughly the slowest source, and each source times out silently after 5 s
The same keyword and source reuse a cached result for 5 minutes, so switching categories does not re-call everything
Files never joins an aggregate: it queries your local SQLite index and its rows have no URL or thumbnail
AI is a conversation, not a retrieval source, so it is excluded too
Music results play inline; a Navidrome hit is your own track and plays on click
Diagram: the category decides who gets asked, each source times out on its own, early results render first, then everything merges.
With no web engine configured at all the Aggregate button does not appear, but the six quick-jump engines still work — they are plain links, key-free and free of charge. Custom search sources (URL template + parsing rules) are a Pro feature.
Search Your Own Media Libraries Too
Configured integrations register themselves as searchable sources — nothing extra to set up. Each one works as a single source or inside an aggregate, and the results come from your own library, never through a third party.
Jellyfin / Emby → Movies: matches films and series in your library, poster straight to the detail page
Navidrome → Music: matches your own tracks and plays them in the panel
Immich → Images: matches photos in your own albums
The test is whether a service returns structured entries with thumbnails for a keyword — status-only services (monitors, download queues) are not stuffed into search results
If a library returns nothing, check that the integration is enabled with valid credentials. A source marked unresponsive in an aggregate is just its own 5 s timeout and does not hold up the others.
Custom Search Sources
Beyond the built-in engines, Pro users can plug in any search API as a custom source. Configure it in Search → API Settings → Custom Sources.
Blank template: configure any API endpoint, params and result fields from scratch
Preset templates: ready-made templates for docs / forum search
Smart fill: with AI configured, auto-analyze the API and fill field mappings
Field mapping: define title / link / thumbnail fields, with regex extraction
Test the connection before saving
Custom sources join aggregates as well — they run concurrently with the other sources of their category
Custom sources are self-configured — you are responsible for the legality of the source and content compliance. Obtain any required API keys through official channels.
Local File Search
The Files category builds a full-text index (SQLite) over shared directories on your server — search file names and contents, click to open. Index and files never leave your device.
Bookmark Jump
The bookmark card supports search and collapsible groups; type part of a title to fuzzy-match and open it. Imported browser bookmarks sync into the Favorites category.
Pinyin works too: typing xunlei or xl matches a bookmark named 迅雷 — full spelling and initials both work
Mixed Chinese/English input, case-insensitive, with the matched fragment highlighted
The search covers group names and site descriptions, not only titles
Press Enter to open in a new tab, or hold a modifier to open in the current page
Pinyin matching runs entirely in your browser — nothing is sent to any service.
AI Answers
Configure your own OpenAI-compatible endpoint (baseURL + apiKey + model, local Ollama works) in the Pro panel AI settings, then ask in natural language from the AI category. Requests go straight to your endpoint — never through BestNav servers.