Nouvelles fonctionnalités, améliorations et mises à jour de Knowbase.
Thinking Mode now uses document summaries to guide its retrieval strategy — the AI reads your file summaries first, then decides which documents to search deeply. This dramatically improves answer quality for cross-document questions. We’ve also upgraded to the latest generation of models across the pipeline.
New users are now greeted with a welcome modal and an interactive step-by-step tour that walks through uploading files, chatting, and sharing. The embeddable chat widget now supports server-driven configuration — change appearance, colors, and behavior instantly without re-deploying your embed code.
Build and deploy custom AI assistants powered by your documents. The full-featured Assistant Hub includes:
Import content from any website directly into your knowledge base. Provide a URL and Knowbase crawls the pages, extracts the text, and makes it searchable in your library and chat conversations. Pages that can’t be extracted are gracefully skipped.
Thinking Mode has been rebuilt with an agentic architecture. The AI now autonomously searches your documents multiple times, assesses its confidence, and synthesizes a comprehensive answer. A Knowledge Mode badge shows when deep analysis is active, and sources are filtered to only those actually cited in the response.
The library has been redesigned with document thumbnails, a compact card layout, and inline file renaming. In Chat-All, clicking a source citation now opens an inline preview panel — view PDF pages, transcript segments, or document excerpts without leaving the conversation. We’ve also added PPTX OCR processing and table-dense PDF detection.
Upload images (PNG, JPG, and more) directly to Knowbase. Text is extracted using our OCR pipeline and made searchable alongside your other documents. Pro and Team plans get fast OCR processing powered by Mistral.
Audio and video transcriptions now include automatic speaker identification. Rename speakers, filter chat queries by speaker, and export subtitles in SRT or VTT format. Speaker labels are preserved in citations so you always know who said what.
Introducing new pricing tiers — Free, Starter, Pro, and Team — with a free Pro trial for new users. Each plan has clear limits for files, queries, and features. The app now includes feature gating so you always know what’s available on your plan, plus a full-width chat layout for a more focused experience.
Knowbase now supports 10+ additional audio and video formats including MOV, WebM, WAV, M4A, AAC, OGG, FLAC, AVI, and MKV. You can also record voice notes directly in the upload view — record, transcribe, and chat with them instantly.
Embed an AI chatbot on your website powered by your Knowbase documents. The widget supports bubble and full-panel modes, customizable appearance, and visitor feedback collection. Copy the embed code and paste it into your site — it just works.
Connect your Dropbox account to import files into your library and search Dropbox content directly from chat. Files from Dropbox are indexed and searchable just like any other source in your knowledge base.
Connect your Notion workspace and search pages directly from Chat-All. Knowbase uses intelligent keyword extraction and title reranking to find the most relevant pages, with Redis caching for instant follow-up queries.
Talk to your documents with voice input. Press Cmd+K (or tap the mic button on mobile) to start dictating your question. Recording auto-stops after 3 seconds of silence and sends your message automatically.
Connect your Google Drive to import files into your library using Google Picker, or toggle Drive Search in Chat-All to search your Drive files in real-time alongside your uploaded documents.
No more waiting for files to finish processing. Drop a file into the chat and start asking questions immediately — Knowbase extracts text on the fly and uses it as inline context. When processing completes, the file seamlessly transitions to full vector search.
The entire application has been redesigned with a modern interface:
For complex questions, toggle Thinking Mode and watch the AI reason through your documents. It performs multiple rounds of retrieval, evaluates its confidence, and reflects before generating a final answer. Watch the thinking steps in real-time as it analyzes, searches, and synthesizes.
Chat responses now stream in real-time with inline source citations. Behind the scenes, we’ve rebuilt our search engine — combining semantic vector search with keyword matching and Cohere reranking for dramatically better retrieval. Document processing has been upgraded with enhanced PDF extraction and multilingual support.