AI Chat

Talk to your documents: pick a model, attach files, receive streaming answers with page citations, track agent progress per step — then ask AI to draft documents, data tables, and slide outlines with previews resembling the originals, revisable through follow-up chat and downloadable.

URL: https://www.documind.id/dashboard/chat

Quick Summary

The Chat Page

URL: https://www.documind.id/dashboard/chat

Page parts:

1. AI model picker — switch models (e.g. mimo, GPT): light models for quick questions, large models for long-document analysis.
2. Chat box + slash commands — type questions or use / for quick commands (summarize, extract, compare). Slash commands beat typing long instructions.
3. Document attachments — attach files from My Documents so answers ground on file context, not generic model knowledge.
4. Streaming answer area — answers stream via SSE (word by word) with clickable [D1], [D2] citations jumping to source pages. Click every citation to verify — never quote answers without checking them.
5. Session sidebar — every session auto-saves. Deleting a session always asks for a confirmation dialog first. Name sessions clearly (topic + date) when juggling many documents.

How to use chat:

1. Pick an AI model in the model picker.
2. Type a question or use / slash commands.
3. Attach documents from My Documents when file context is needed.
4. Read the streaming answer; click [D1], [D2] citations to verify against source pages.
5. Follow up with narrowing questions ("focus on Chapter 3", "compare with appendix B").

Good vs weak questions:

• Weak: "Explain this document." (too broad, generic answer)
• Good: "Summarize Chapter 3 in 5 bullets + cite the source page per bullet."
• Good: "Extract every budget figure in appendix B into a table: item, value, page."
• Good: "Compare definition X in document A vs document B and highlight differences."

Agent Task Detail

URL: https://www.documind.id/dashboard/task/:sessionId

For heavy jobs (long summaries, complex extraction, research), chat runs as a multi-step agent task. Page parts:

1. Step list — each step shows operation name and status (running/done/failed).
2. Live running duration — each step shows its running duration (e.g. thinking 51s) computed live from backend timestamps (now - timestamp), stopping when the next step starts. You see exactly where time goes.
3. Per-step output files — view and download each step's result (not just the final), useful for audits and partial reuse.
4. Stable workspace — the sess-<session_id> working folder is never auto-deleted; only cleaned when untouched for > 1 week via cron. Come back anytime, files remain.
5. Error masking — technical errors are masked; users see only friendly messages, never provider stack traces. On failure, retry the step or simplify the request.

Document Drafts from Chat

The biggest time-saver: beyond answering, chat can draft content rendered as a preview resembling the original look — then you revise through follow-up chat ("add a methodology section", "make it more formal", "shorten to 1 page") and download the result.

Document Preview (Original-Like Look)

Ask in chat: "Draft a 2-page executive summary of this document" or "Write an offer letter based on the attachment". The result renders as a document preview resembling the original appearance — headings, paragraphs, lists, and tables reading like a word processor — not raw chat text.

Preview-area parts:

1. Preview body — tidy render: title, nested headings, justified paragraphs, bulleted/numbered lists, clearly-ruled tables.
2. Structure navigation — for long drafts, headings act as a map for jumping between sections.
3. Revision commands — continue in the same chat ("switch the greeting to formal", "add a comparison table in Chapter 2"); the preview updates without restarting from zero.
4. Download & copy — copy contents or download as a file; polish conversions with Document Converter, Text to DOCX, or Markdown to PDF when a specific format is needed.

Uses: meeting notes into formal minutes, papers into 250-word abstracts (check with Word Counter), CV and cover-letter drafts, weekly reports from rough notes.

Table & Data-Grid Preview (Spreadsheet-Like)

Ask in chat: "Extract every invoice line into a table: no, description, qty, price, total" or "Compare these 5 papers: title, year, method, findings". The result renders as a spreadsheet-like data grid — headed columns and scannable rows, not paragraphs.

Its parts:

1. Preview grid — headed columns, data rows, right-aligned numbers for easy comparison.
2. Completeness check — ensure no cell that should be filled is empty; follow up ("fill the empty year column").
3. CSV download — take results to a spreadsheet for further work: open with CSV Viewer, analyze with CSV Statistics, merge with CSV Merge, convert with CSV to JSON or JSON to CSV as needed.

Uses: monthly invoice recaps, survey-result tabulation, product-spec comparisons, structured bibliographies.

Presentation-Outline Preview (Slide-Order-Like)

Ask in chat: "Build a 10-slide presentation from this report" or "Draft opening + body + closing slides for the briefing". The result renders as a per-slide preview resembling original deck order — slide titles, bullets, notes — so the story arc can be judged before moving to a presentation app.

Its parts:

1. Slide list — each slide clearly titled with 3–5 bullets (readability rule of thumb).
2. Story arc — opening (problem/goal) → body (data/arguments) → closing (recommendations/actions). Reorder via chat ("move key findings to slide 2").
3. Copy to presentation — copy slide by slide into your favorite presentation app; adjust theme and visuals there.

Uses: policy-briefing slides, research-proposal pitches, monthly recaps for leadership.

Real-World Flows

Student dissecting 3 journals (45 min): attach 3 PDFs → "summarize each in 5 bullets + source pages" → "compare their methods in a table" → download the table → verify citations one by one via [D1] → save papers to My Library.

Admin recapping invoices (20 min): attach 12 invoice PDFs → "extract into a table: no, date, vendor, total" → download CSV → open in CSV Viewer → total up and archive in Library.

Job seeker (30 min): paste raw experience → "draft a 1-page CV + formal cover letter" → revise via chat ("emphasize experience X") → download → length-check with Word Counter → finalize with Text to DOCX.

Tips & Best Practices

1. Specify output format up front. "Answer in a table / 5 bullets / 1-page draft" yields one-shot results without wasteful iterations (Billing).
2. Attach only what is relevant. 2–3 focused documents beat 20 mixed files — sharper answers, fewer credits.
3. Verify every important citation. Click [D1], [D2] before citing into papers/official reports.
4. Revise incrementally, never restart. "Shorten section 2" is cheaper and more consistent than requesting a fresh draft.
5. Name sessions clearly. Thesis-Ch2-2026-11-01 beats New Chat in the sidebar.
6. Separate exploration vs final chats. A free chat for digging; a clean new chat for the final draft, keeping context unpolluted.
7. Finish with tools. Count words (Word Counter), convert formats (Document Converter), tidy markdown (Markdown to HTML) — all free.

Limits to Know

• Every model call spends credits per tokens — track in Billing.
• Models can err (hallucinate): citations exist to be verified, not decorated. For critical claims, always check source pages.
• Technical errors are masked into friendly messages — on failure, simplify or split into smaller steps.
• Session workspaces are cleaned when untouched for > 1 week — download important files to Library before that.

AI Chat FAQ

Which model should I pick? Light ones for quick questions and short drafts; large ones for long documents and complex analysis. When unsure, start with the default.

Can chat read many documents at once? Yes, attach several files. Practical limit: more context costs more and generalizes answers — prioritize relevance.

How do I ensure accurate answers? Click every [D1], [D2] citation and read the source page. Ask the AI to cite pages for each key claim.

Can chat outputs be downloaded? Yes — document drafts, tables (CSV), and slide outlines can be copied/downloaded from their preview areas.

Deleted session — recoverable? Deletion asks for confirmation first. Once confirmed, sessions are unrecoverable — download important outputs to Library habitually.

Does chat spend credits? Yes, per processed tokens. Do deterministic work (convert/count/format) in the free Tools.

Important Links

• Overview · Authentication · Account · Billing · Landing Page
• Dashboard · AI Chat · Research · Find Journal
• Library · My Library · Tools

Keywords: documind ai chat, ask documents, page citations, summarize documents, ai document draft, extract tables, presentation outlines.