Open a citation to see a format-aware preview of its source. PDFs and Office files supported by the server's LibreOffice renderer can be viewed page by page; other files use the matching table, media, or text preview when their content is available. Large CSV files load in bounded row pages so the browser does not need to display the whole file at once. If a visual renderer is unavailable, Query reports that and keeps the available format preview accessible.

CSV and TSV attachments use the same saved, security-scanned file in Documents Hub. If an accessible copy already exists, Query attaches it by its document ID without uploading a duplicate. After the normal Documents Hub indexing finishes, a separate Grounded Query worker builds a complete table index and shows its stage and progress in the composer.

Grounded Query accepts tables up to 25 MiB, 1,000,000 data rows, and 256 columns. It rejects malformed or over-limit files instead of silently leaving rows out. Ask for calculations such as counts, sums, averages, filters, sorting, or grouped results. Those operations run against the full indexed table; retrieval supplies explanations and source context. The model that plans a table operation receives the schema, not every table cell.

Answers cite the file and show the matching-row count plus source-row examples. Open a citation to inspect the table, page through 100 rows, or jump directly to a source row. Row 1 contains the column names.

1. Make sure the relevant file has completed processing in Documents Hub.
2. Ask one clear question and name a document, date, or subject when that helps narrow context.
3. Check the cited passages and ask a follow-up if the evidence is incomplete or ambiguous.
4. For work that needs broader research, drafting, analysis, notes, or tools, continue in DeepSpace.

1. Search the model menu and choose an available chat model under its provider. Each provider row shows its icon, and opening another provider closes the current one.
2. Use the paperclip to upload files. They follow the normal Documents Hub security and indexing process. Uploads continue while you move between pages in the same open app window or browser tab. Documents Hub shows the in-progress upload live, then replaces it with the saved document and its normal indexing status after acceptance. The temporary upload row is not restored after a full page reload. Wait until each file shows Ready before sending your question.
3. Attachments are included with the selected document or Qel Connect scope, and answers continue to use the existing grounded retrieval and citation path.
4. While an answer is streaming, send another prompt to add it to the session queue. You can pause or resume the queue, remove queued prompts, or stop the current answer. Queued prompts are limited to 20. If an answer fails, the queue pauses and keeps that prompt available to retry before later prompts continue. Queued prompts are cleared when you switch conversations and are not saved across page sessions.

Removing an attachment from the composer only detaches it from that query. The uploaded source remains in Documents Hub, where its existing retention and deletion controls apply.

Answer quality depends on the source content, extraction/OCR, indexing, retrieval settings, and configured model provider. A citation shows what the answer references; it does not prove that the source itself is correct or that the response is complete. Provider and search failures may make some paths unavailable. CSV and TSV files above the published size, row, or column limits, or with malformed rows, are rejected and must be corrected before Query can use them.

Query is optimized for retrieval-first questions over source documents. DeepSpace is a separate broader conversation workspace with its own working Library and optional research, sandbox, schedule, memory, and connector capabilities. Documents Hub remains the home for source document lifecycle and organization.