Your organisation already has the answer. It is spread across ten thousand files nobody can read at once.
Archives nobody queries
Years of contracts, reports, policies and correspondence sit in storage because reading them is not a realistic task.
Search returns documents, not answers
You get a list of files and start reading. The work has been moved, not done.
Assistants that see a few pages at a time
General AI tools take what fits in a context window and quietly ignore the rest.
Answers you cannot defend
A confident paragraph with no source is unusable in a regulated decision.
Context rebuilt by hand, every time
Each team assembles the same background from scratch for every question.
Projects that begin with a migration
Knowledge initiatives stall for a year while somebody moves twenty years of files into a new system.
What DSX IQ does.
Connect, don't migrate
Point DSX IQ at your file servers and network shares. Documents are indexed where they already live and re-checked on a schedule you set. Nothing has to be copied into a new system first.
Two ways in
Upload documents directly, or let them arrive from a connected share — nothing else to configure. PDFs, Office files, email, scans and images all parse, and OCR with layout analysis brings in the documents that were never machine-readable. Every file is segmented, embedded and indexed with its structure and metadata intact, so hierarchy, position and provenance survive into the answer.
Simulated unlimited context
A retrieval and compression layer assembles, ranks and re-ranks exactly the material a question needs, then holds it in a rolling working memory. The model behaves as though it had read everything — with no context window ceiling.
Conversation with memory
Follow-up questions inherit the thread. Narrow, compare, challenge and refine across a long session without restating what you already asked.
Cited, auditable answers
Every claim carries a link to the document, page and passage it came from, with a confidence score and the option to open the source in place.
Governed access
Answers respect the permissions of the person asking. Users see only what they are entitled to see, and every question and response is logged.
No migration. No second copy of the truth.
Most knowledge projects stall before the first question is asked, because they begin by asking an organisation to move twenty years of files somewhere new. DSX IQ starts from the opposite premise: your file servers stay exactly as they are.
Direct connection
File servers and SMB network shares connect through read access. Documents are indexed in place, at the path they already sit on.
Scheduled sync, differences only
Indexing runs on the schedule you set and reads only what moved since the last pass — new files added, modified files re-indexed, deleted files dropped. A share holding two million documents is not re-read every night to find the fifty that changed.
Folder structure as context
The path a document sits in is signal. Client folders, matter numbers, year directories and naming conventions are read as metadata and can be used to scope a question.
Inherited permissions
Existing access control lists are read from the source and enforced at query time. Nobody gains access to a folder by asking a question about it.
Read-only by design
IQ does not write to, move or reorganise your shares.
Connect the share, let it index, start asking. The archive stays where your organisation already trusts it to be.
Connectors for SharePoint, document management systems and cloud drives are in development. Anything not yet connected can be uploaded today.
The whole archive, in a single line of reasoning.
Context windows are a hard limit. A question that touches four hundred documents cannot be answered by a model that reads twenty. DSX IQ removes the ceiling by changing what reaches the model — where retrieval alone runs out explains why.
The question is decomposed into the entities, dates, document types and relationships it actually depends on.
The index is queried across the full corpus. Candidates are ranked, re-ranked and compressed into a dense working set that preserves meaning and drops repetition.
The working set is carried through the answer and across follow-ups. Material is swapped in and out as the conversation moves, so depth on one thread never costs you breadth on the next.
The result: answers that hold across thousands of documents, not just the ones that happened to fit.
Every answer carries its evidence.
Inline citations
Document, page and passage, linked and openable without leaving the conversation.
Confidence scores
Say how firm the ground is, not just what the answer is.
Coverage view
See how much of the corpus informed a given answer, and what was excluded.
Stated gaps
When the record does not contain the answer, IQ says so rather than filling the space.
Enterprise confidence, by default.
One archive, many questions.
Legal
Question a matter's full document set — obligations, precedents, exposure — without reading it end to end.
Financial services
Interrogate credit files, policy manuals and regulatory correspondence in one place.
Insurance
Reconstruct a claim from every document attached to it.
Compliance and audit
Test a control against every policy version and every piece of evidence held.
Engineering and operations
Ask specifications, manuals and incident history as a single body of knowledge.
Independent products. One governed estate.
DSX Insight and DSX IQ are separate products, bought and deployed on their own terms.
Insight works a page at a time
It classifies a document, extracts the fields your systems need, and lets you question that document while it is in front of you. One file, one workflow, one answer.
IQ works across everything at once
It indexes your whole estate and answers questions that span thousands of files in a single conversation.
One is a processing pipeline. The other is a knowledge surface. Neither depends on the other, and most organisations arrive at them through different doors.
Reach your archive from the assistant you already use.
Register DSX IQ in DSX Guard as a governed agent and your organisation's knowledge becomes available inside ChatGPT, Claude and Gemini. An employee asks a question in the tab they already have open; Guard routes it to IQ, applies masking and policy on the way out, and returns a cited answer drawn from your corpus. Every exchange is inspected and logged.
The archive comes to the assistant, not the other way round.
Sensitive values masked before the model sees them, restored in the answer.
Questions, sources and responses recorded in the Guard console alongside every other AI interaction.
A platform base, then priced per user.
DSX IQ sits on the shared DSX platform base and scales on the people using it, from $250 per user a year, with the rate stepping down as that number grows.
Platform
A shared platform base from $25,000 a year, covering deployment, support and maintenance across every DSX product you license.
Users
Per person with conversational access, with volume tiers that step down as the number grows.
Documents in the index are priced separately, per document a year by corpus size, and quoted with your document counts. Ingestion, scheduled re-indexing and retrieval are included. You are not charged per question, so nobody on your team has to ration what they ask.
See pricingQuestions about enterprise knowledge intelligence.
How is this different from search?
Search returns documents and leaves the reading to you. DSX IQ answers the question — including questions whose answer is spread across hundreds of documents and exists in none of them alone — and cites the document, page and passage each part of the answer came from.
Do we have to move our documents?
No. IQ connects to the file servers and SMB shares your teams already use and indexes the estate in place. Re-indexing is scheduled and difference-only, so only what changed is reprocessed, and files removed from a connected share drop out of the index at the next sync.
Can people see things they should not?
Retrieval is permission-aware: access control lists are read from the source systems and enforced at query time, so an answer is composed only from documents the person asking is entitled to see. Who asked what, and what was consulted, is recorded.
What about scanned documents?
OCR and layout analysis are part of ingestion, so scanned and image-only material is indexed and answerable alongside born-digital files.
How does it handle questions bigger than a context window?
A retrieval and compression layer assembles, ranks and re-ranks exactly the material a question needs and holds it in a rolling working set, so a question touching hundreds of documents is not limited to whatever fits in one prompt. Each answer also shows how much of the corpus informed it and what was excluded.
Does every question cost us money?
No. Ingestion, scheduled re-indexing and retrieval are included, and questions are never metered. Licensing is per user with a per-indexed-document rate alongside it.
Does IQ bypass our AI policy?
No — it registers as a governed agent in DSX Guard, so the model calls IQ makes pass the same policy layer as every other path to AI in the organisation.
Where does it run?
Cloud, private cloud or on-premise, with local model options where data residency requires it.