Q

2.0

Products DocIQ Chat
💬 DocIQ Chat — Q&A on your documents Page-level Citations Grounded retrieval

Every answer, with the receipt.

Ask your policies, contracts, manuals, filings. Get answers grounded to the page.

DocIQ Chat is the chat interface for your unstructured documents — PDFs, Word, HTML, wikis, SharePoint, Box. Retrieval-Augmented Generation with page-level citations and source-text snippets. Per-domain agents already trained on your taxonomy. Never a plausible-sounding guess. Always a passage you can defend in writing.

Built for analysts, legal, compliance, caseworkers — anyone who needs the answer and the paragraph it came from. No hallucinations. No misquoted regulations. No "source: trust me."

💬
DocIQ Chat
Q&A with citations · per-domain agents
Indexed
🏛 Retirement Benefits Agent · 24 docs
👤
What is the process for evaluating post-retirement earnings limit in a benefit-determination case?
For Post-Retirement Earnings Limit in a benefit-determination case, follow SOP-PB-041. Review earnings across Plan-Covered, External Wage Authority, and Member-Reported sources, then trigger Perform Earnings-Cap Evaluation in MemberServices2. If earnings exceed the calendar-year limit, reduce benefit liability proportionally3.
1
Benefits_Eligibility_SOP.pdf · p. 4, §3.1
"...the evaluation follows the SOP-PB-04 procedure for all benefit-determination cases involving Earnings-Limit Coverage..."
2
MemberServices_Guide.pdf · p. 12
"Click Perform Earnings-Cap Evaluation to launch the BPM..."
3
Benefits_Eligibility_SOP.pdf · p. 7, §4.2
"...reduce benefit liability proportionally to earnings over the cap amount..."
Ask a question…
DocIQ Chat in production
100%
Cited Answers
<2s
Retrieval Time
6+
Domain Agents
8+
Document Types
0
Hallucination Tolerance
VPC
Private Deploy
Why DocIQ Chat Exists

A chatbot that can show its work.

Most enterprise chatbots hallucinate. They sound confident, they quote convincingly, and they make up case numbers, clause references, and page citations that don't exist. In legal, compliance, clinical, and policy work, that's worse than no answer — it's a liability-generating answer.

DocIQ Chat refuses to generate what it can't cite. Every claim maps back to a page, a paragraph, a filename. If the corpus doesn't contain the answer, it says so — and helps you narrow the search instead of inventing text.

Confident without citation is a liability. DocIQ Chat cites everything.

Domain Agents

Six experts. One chat box.

DocIQ Chat ships with pre-configured domain agents. Each one is scoped to its own corpus and trained on its own taxonomy. Switch with one click.

🏛
Retirement Benefits
Pension · Annuity · Survivor
benefit-determination cases, Earnings-Limit Coverage, earnings over the cap, SOPs, BPA workflows. The flagship agent at our pension reference deployment.
24docs
1,247chats
4.8★
Compliance Advisor
GASB · DOL · Regulatory
Government code, agency rules, regulatory filings. Ask "what's the deadline for X?" — get the section number.
18docs
892chats
4.6★
👥
HR Policy
Policy · Leave · Benefits
Employee handbook, leave policies, benefits enrollment. Self-service Q&A for HRIS without a ticket.
15docs
634chats
4.5★
📜
Legal & Contracts
Contract · Compliance · Legal
Contract clauses, indemnity caps, termination terms. Find every contract that mentions a specific obligation.
9docs
412chats
4.4★
📊
Finance & Reporting
Budget · Forecast · Investment
Budget docs, forecasts, investment reports. Ask the financials and get the page that explains the variance.
12docs
578chats
4.7★
💻
IT Helpdesk
Runbooks · KB · Support
Internal runbooks, knowledge base, support workflows. Tier-1 self-service that actually deflects tickets.
6docs
234chats
4.3★
What It Reads · Where It Reads From

Every type of document. From everywhere they live.

DocIQ Chat doesn't care what format your knowledge is in or which silo it's stuck in. It indexes, embeds, and retrieves from all of them — equally, with one search bar.

Every type of document
📕
PDFs
.pdf
Contracts, policies, state codes, filings, manuals. OCR for scans, native text for clean exports.
📘
Office Docs
.docx · .pptx · .xlsx
Word, PowerPoint, Excel with text. Full-fidelity parsing of structure, tables, and footnotes.
🌐
HTML & Web Pages
.html · public sites
Public HTML, intranet pages, knowledge-base articles. Structure-aware chunking by heading hierarchy.
📋
Markdown & Notes
.md · .txt · Notion exports
Internal runbooks, engineering notes, exported wikis. Heading-aware retrieval; preserves code blocks.
📧
Email & Threads
.eml · .msg · .mbox
Email archives, exchange exports, discussion threads. Thread-aware retrieval with sender attribution.
🎙️
Transcripts
.vtt · meeting notes
Meeting transcripts, call logs, customer interviews. Time-coded retrieval with speaker attribution.
🖼️
Scanned Images
.png · .jpg · OCR
Scanned letters, forms, receipts, certificates. OCR with confidence scoring surfaces only well-read text.
💾
Code & Config
.py · .yaml · .json
Source code, infrastructure config, schema files. Useful when engineering wikis are out of date.
From wherever they live

Box AI works on Box. Copilot works on SharePoint. DocIQ works on all of them.

Connect once, sync forever. Permissions inherited from the source. Your bytes never leave your storage.

Amazon S3
object storage
Azure Blob
storage
Google Cloud
storage
SharePoint
+ Teams
OneDrive
M365
Google Drive
Workspace
b
Box
content cloud
Confluence
+ wikis
Drag & Drop
local upload
Connect once, sync forever. Set a cadence — hourly, daily, on-change. DocIQ ingests incrementally, honors source ACLs, and keeps the index in sync with deletes, renames, and new versions. Network shares, FTP, and custom connectors also supported.
What DocIQ Chat Does

RAG that actually works at work.

🎯
Page-level citations, always
Every factual claim cites the exact document, page, and paragraph. One click opens the highlighted source passage.
  • Document name + page number
  • Highlighted passage preview
  • Exportable citation bundle (DOCX / PDF)
🧩
Hybrid retrieval, not vibes
BM25 for keyword precision, pgvector embeddings for semantic recall, reciprocal-rank-fusion to combine. Beats pure-vector on every benchmark.
  • BM25 + pgvector + RRF fusion
  • HNSW index · 1536d embeddings
  • Query re-writing for retrieval
🚫
"I don't know" is a feature
If retrieval confidence is below threshold, DocIQ Chat refuses to generate. Asks the user to rephrase or points at closest documents. No fabrication.
  • Tunable confidence threshold
  • Out-of-corpus detection
  • Guided rephrase assistance
🔄
Versions & effective dates
Knows the 2019 policy from the 2025 amendment. Retrieval defaults to the latest effective version; historical is explicit.
  • Document version tracking
  • Effective-date awareness
  • Time-travel queries
🔐
Access enforced at retrieval
A user can only chat with the documents their role can see. Access-control lists honored at retrieval time — no information leaks through summaries.
  • ACL-aware retrieval
  • Honors source-system permissions
  • Full audit log per chat turn
🏢
Runs in your VPC
Private VPC, air-gapped, or on-prem. Documents never leave the boundary. Embeddings computed locally. LLM gateway isolated.
  • Air-gapped deployment
  • Local embedding & inference
  • Zero-retention configurations
How It Works

Ingest. Retrieve. Ground. Cite.

Four stages, one loop — every generated word traceable back to a source passage.

1
📥
Ingest
Parse, OCR where needed, chunk by semantic structure, embed with text-embedding-3-small (1536d).
2
🔍
Retrieve
Hybrid BM25 + pgvector search with RRF fusion. Top-k candidate passages with confidence scores.
3
🧠
Ground
LLM generates only from retrieved passages. Unsupported claims are stripped; low-confidence answers are refused.
4
📎
Cite
Every claim linked to document, page, and passage. Exportable citation bundle travels with the answer.
Citations · receipt attached

Click a footnote. See the paragraph.

DocIQ Chat doesn't just name a source — it shows you the passage. Every footnote opens a preview panel with the highlighted text, the document name, the page number, and a deep link to the original.

When a regulator, auditor, or executive asks "how did you arrive at this answer?" — the answer is already attached.

  • One-click preview of cited passage
  • Deep link to source document
  • Exportable citation bundle for compliance
  • Full chat audit log — every answer, every source
Benefits_Eligibility_SOP.pdf · Post-Retirement Earnings Limit p. 7 / 24

4.2 Earnings-Cap Treatment

When a member's reported earnings exceed the calendar-year limit defined in the MemberServices Earnings-Limit Table, the analyst shall complete the evaluation worksheet attached as Appendix B.

If the cumulative earnings across Plan-Covered, External Wage Authority, and Member-Reported sources exceed the threshold, reduce benefit liability proportionally to the earnings over the cap amount, and document the calculation in the case notes field within MemberServices.

Refer to the SOP-PB-04 decision matrix for edge cases involving partial-year reporting or retroactive earnings adjustments...

Doc: Benefits_Eligibility_SOP.pdf ✓ retrieved · RRF score 0.91
Use Cases

Where DocIQ Chat earns its keep.

⚖️
Contract & Clause Search
Legal · procurement · commercial
Ask "which of our contracts has an indemnity cap below $5M?" — DocIQ Chat searches the portfolio, returns every match with clause number, page, and excerpt. No intern week required.
✓ Contract questions answered in seconds, with receipts
🏛️
Policy & Regulation Navigation
Government · state agencies · regulators
State code, agency rules, federal registers. Staff ask "what's the reporting deadline for X?" — DocIQ Chat cites the exact section. Caseworkers stop misquoting regulations.
✓ Citizen-facing answers defensible to the AG
⚕️
Clinical Policy & Protocols
Payers · providers · HIM
Utilization managers and coders chat with coverage policies, clinical guidelines, and coding manuals. Every determination cites the paragraph. Appeals defensible.
✓ Coverage decisions with audit-ready citation trails
🛡️
GRC & Audit Evidence
Compliance · internal audit · risk
Ask "what controls do we have for SOC 2 CC6.1?" — DocIQ Chat returns policy text, procedure references, and evidence excerpts with page-level citations. Audit prep in hours, not weeks.
✓ Audit evidence packets assembled from natural-language queries
Why DocIQ Chat

Not a corporate chatbot.

DocIQ Chat is a grounded-retrieval product purpose-built for enterprises that cannot afford a hallucinated answer.

Federated — not vendor-locked
Box AI works on Box. Copilot works on SharePoint. DocIQ Chat works on every store at once — S3, Azure, GCS, SharePoint, OneDrive, Drive, Box, Confluence. One question, every silo.
Zero hallucination tolerance
If a claim can't be grounded to a retrieved passage, it's stripped from the answer. Low-confidence questions get refused, not fabricated.
Hybrid, not pure vector
BM25 keyword + pgvector embeddings + RRF fusion. Finds exact phrasing and semantic neighbors — beats either one alone.
Citations users actually trust
Not "sourced from N documents." The document name. The page. The highlighted paragraph. One click to the original.
Source ACLs honored
Retrieval respects the source system's access controls. A user never sees — or summarizes — a document they shouldn't have access to.
Private VPC · air-gapped
Documents never leave the boundary. Embeddings and inference run in your infrastructure. No data training anyone's model.

Chat with your documents — with the receipts attached.

See DocIQ Chat answer a question from your corpus of policies, contracts, or filings — grounded, cited, page-level accurate. Thirty-minute demo, your documents.