CorpusIQ is now live in the ChatGPT app store.
Find CorpusIQ in ChatGPT
Connector directory
Files & Storage
Advertising & Analytics
CRM
Email Marketing
Commerce
Finance
CorpusIQ is now live in the ChatGPT app store.
Find CorpusIQ in ChatGPTConnector directory
The CorpusIQ Intelligence Layer is the set of 12 engines that sit between your business tools and the AI asking the question. They work in three stages. Understand what is really being asked. Validate every number involved. Apply the right business analysis and cite the source. The assistant never touches your raw data directly, which is why the same question returns the same number every time.
Read-only OAuth · Zero customer file storage · Source-cited answers
Connect an AI assistant straight to your business tools and it will answer every question you ask. That is the problem. A language model handed raw rows produces a fluent answer whether or not the numbers reconcile, and it has no way to know that two systems disagree about the same order, or that your definition of churn excludes trials.
Ask the same question twice and you can get two different numbers, both delivered with total confidence. Ask it in a different assistant and you can get a third. Nothing in the model is doing the work of deciding which system owns revenue, or what your company means by an active customer.
The Intelligence Layer does that work first. By the time the assistant sees anything, the question has been resolved into a plan, the entities have been matched across systems, the definitions have been pulled from a registry, and the figures have been checked. The assistant narrates a validated result instead of improvising one.
This page describes the CorpusIQ implementation. For the category itself, and how an intelligence layer differs from a connector platform, an ETL pipeline, or a BI dashboard, see what an AI intelligence layer is.
Grouped into the three stages every question passes through, in order.
Works out what you are really asking and what your data means.
Turns your question into a structured AI workflow.
A question like "how did we do last month" is not a query. This engine turns it into an ordered plan: which systems to read, which date range applies, which records matter, and in what order the steps have to run. The plan is what executes, not the sentence.
Works out what you are really asking, even when the question is vague.
Most business questions are underspecified. "Best customer" might mean highest revenue, highest margin, or longest tenure. This engine resolves the ambiguity against how your business actually defines the term, and says which reading it used, so you are never guessing what the answer measured.
Understands how customers, products, campaigns, and orders relate.
Knowing that a Shopify order, a Stripe charge, a HubSpot deal, and a Gmail thread all belong to the same relationship is what makes a cross-tool answer possible. This engine holds those relationships so a question can cross systems without you naming every join.
Matches the same customer, product, or company across multiple systems.
The same company is Acme Inc in the CRM, ACME in the accounting file, and acme.com in the analytics. Entity Resolution matches them so revenue does not get counted three times under three names, which is the most common way cross-tool numbers go wrong.
Checks every number is correct and consistent before the AI sees it.
Keeps KPI definitions consistent across the whole company.
Define a metric once and it stays defined. If churn excludes trials and counts on the renewal date, every answer uses that rule, in every assistant, for every person on the team. This is what makes the same question return the same number twice.
Uses the correct system of record for every metric. No guesswork.
Revenue comes from the accounting system, not the ad platform. Sessions come from analytics, not the CRM. This engine holds which system owns which metric, so a number is never quietly sourced from whichever tool happened to answer first.
Verifies every number before it reaches the AI. No validation, no answer.
Figures are checked before the model ever sees them. Totals have to reconcile, date ranges have to line up, and records have to be complete. If a number cannot be validated, CorpusIQ says so rather than passing an unverified figure to the assistant to narrate confidently.
Stops the AI inventing metrics, redefining KPIs, or changing rules.
Language models are fluent enough to redefine a metric mid-answer and sound right doing it. This engine pins the definitions so the assistant reports the number it was given under the name it was given, instead of quietly substituting a plausible calculation of its own.
Detects missing, conflicting, or suspicious data before analysis.
Gaps and contradictions surface before they become a conclusion. A missing week of ad data, a duplicated invoice, or two systems disagreeing on the same order gets flagged in the answer rather than silently averaged into it.
Applies business skills, reads your documents, and shows its sources.
Applies business-specific skills across finance, marketing, and operations.
Skills are pre-built analyses that know what a good answer to a given business question looks like. Asking about overdue invoices runs the collections logic; asking about ad efficiency runs the attribution logic. You get the analysis, not just the raw rows.
Reads PDFs, invoices, contracts, statements, and more.
A lot of the answer lives in documents rather than databases. Invoices, signed contracts, and bank statements get read as data, so a question about payment terms can reach the actual clause instead of stopping at what someone typed into a field.
Shows exactly where every answer came from.
Every figure carries a link back to the record it came from. That is what makes an answer checkable by someone who was not in the conversation, and it is the difference between a number you can take to a board meeting and one you cannot.
Define a metric once and it holds. Ask in ChatGPT, Claude, or Perplexity and the arithmetic is identical, because the same registry answered all three.
Every figure links back to the record behind it, so the answer survives a question from someone who was not in the conversation.
The engines run at the moment of the question against 40+ connected tools. No index, no standing replica, no training on your data.
150+ skills carry the logic for what a good answer to a given business question looks like, so you get the finding rather than the export.
It is the set of 12 engines that sit between your connected business tools and the AI assistant asking the question. The engines work in three stages: understand what is being asked, validate the numbers involved, then apply the right business analysis and cite the sources. The assistant never reads your raw data directly.
Because a language model handed raw rows will produce a fluent answer whether or not the numbers reconcile. It has no way to know that two systems disagree about the same order, or that your definition of churn excludes trials. The layer resolves those things before the model sees anything, so the assistant is narrating a validated result rather than improvising one.
Because the definition does not live in the prompt. The Metrics Specification Registry holds it, the Source of Truth Engine decides which system owns it, and Anti-Drift Protection stops the model substituting its own calculation. Ask in ChatGPT on Monday and Claude on Friday and the arithmetic is identical, because the same registry answered both.
No. There is no index and no standing replica of your corpus. The engines run at the moment of the question, pull only the slice that question needs through read-only OAuth, and retain nothing afterward. Customer data is not used for model training.
Yes. Audit and Citations attaches the source record behind every figure, so an answer can be checked by someone who was not in the conversation. That traceability is the point; a number you cannot trace is a number you cannot take into a board meeting.
No. They run by default on every plan. The one piece worth configuring is your metric definitions, so the registry matches how your business actually counts things. Everything else works from the moment a connector is authorized. 30-day free trial.
30-day free trial. Read-only OAuth. Cancel anytime.