Grounded Query
AverQel still has a strong grounded retrieval layer for document-first work: upload, parse, chunk, embed, retrieve, stream, and cite source-backed answers.
Document Pipeline
Uploaded files are processed into text, chunked for retrieval, embedded, indexed, and tracked with processing status, progress, and extraction metadata.
Grounded Answers
The query runtime is built to answer from accessible documents and return results tied to actual source material instead of only ungrounded generation.
Rich Output
The query UI supports markdown, charts, diagrams, and structured blocks so grounded results can be presented as more than plain text.
Document Inspection
Users can inspect status, full text, versions, chunks, download the original file, and save extracted or selected content into DeepSpace notes.
What the user-facing document system includes
- document uploads with live processing progress
- supported format discovery
- download and full-text viewing
- reingest and retry support
- quarantine/extraction quality signals
- query page for grounded question-answering
- save-to-note flows that turn source material into DeepSpace workspace content
Document Hub file and security contract
The Hub advertises its extractor registry at upload time. Native support includes PDF, TXT, Markdown, OCR-capable images, DOCX, PPTX, XLSX, and the configured code/text formats. Legacy DOC, PPT, and XLS files are converted through the server-side LibreOffice worker when that feature is enabled. Arbitrary binary files, archives, audio, and video are rejected.
Every upload is checked against the configured extension and detected MIME type, archive safety limits, and the production ClamAV service before it enters object storage. The original file can be downloaded from the detail view; browser rendering is limited to safe PDF and image previews, while other binary formats are downloaded for native opening.
How it differs from DeepSpace chat
Grounded query is best when the user wants evidence-backed answers over documents. DeepSpace is best when the user wants a broader productivity conversation with memory, safe retrieval, and optional source inspection.
Both surfaces are important: Query is retrieval-first, while DeepSpace is the broader conversation surface.
Why it still matters
AverQel depends on solid grounded retrieval to turn private files into usable, trustworthy context for both query answers and DeepSpace conversations.