Asked often, answered plainly.

No. A chat layer writes a query against one schema and returns whatever comes back. Synaptic Intelligence sends a question to every department that holds part of it and gives each business word one meaning, owned by one department, so revenue in the ledger and revenue in the CRM are reconciled as two figures for one thing rather than treated as the same number. It verifies every citation before showing you anything.

Getting an answer is easy. Knowing whether you can trust it is harder. The visible difference shows up when the systems disagree or a term is undefined: most tools pick one and sound certain regardless. This one shows you the disagreement.

You can, and a technical team will try it first. What that gives you is a configuration someone has to own: servers picked by hand, auth maintained, config files updated as APIs change. It gives one agent a long tool list to search, one set of credentials for everyone who asks, and no persistent definition of contested words like employee, so the same question phrased twice returns two numbers.

What it does not give you is a checked source for every figure, per-person access control, or a product a vendor can repeat across a hundred organisations. The connectors are the engine. The onboarding, orchestration, semantic layer, governance and trust surface are the product.

No. BI renders what an analyst modelled in advance; this decides at query time which departments to consult and how to combine them, which is how it answers questions nobody built a dashboard for. Revenue per employee needs a warehouse join in BI and needs no join here.

Where BI is better, we say so: a reviewed SQL view is more auditable than a live computation. In most organisations this sits alongside BI rather than replacing it.

Yes, along two separate axes. An access role (viewer, member or admin) decides what a person may change: who may connect a system, who may edit the company profile, who may spend money asking a question. A business title decides which departments a department may read on their behalf, and travels into the prompt every department works under.

Neither is something a person can set for themselves. Both are checked again behind every endpoint, so calling the API directly gets the same refusal the interface implied. Sign-in is Microsoft Entra ID or Google, or a single-use link emailed to someone with no directory, so revoking someone in your directory ends their access from their next sign-in.

Wherever it lives now. There is no warehouse: records are read at the moment a question is asked, used to compose that answer, and not copied into a store of ours. While a conversation is under way, each department keeps what it read for 24 hours after the last question, so a follow-up does not read everything again; then it is deleted. What we record is the question: what was asked, which departments answered and what it cost. The answer, with the record id behind every figure, stays with the person who asked it and exports to a file, so every figure can be traced to its source without us holding your data to do it. Documents are the exception: a file you upload or sync from a drive is held, in passages, so it can be searched. Beside that, each department keeps short notes on how your company works, such as how it counts things and where its figures live: never figures, names or records, and an admin can switch learning off.

Connections are read-only by construction, and the credential you issue is the real boundary. You can run the platform in a particular region, or inside your own cloud account. Your records are not used to train models.

No. Every request is resolved to one company before anything is read, and every query is scoped by it: connectors, documents, definitions, spend and people alike. There is no query in the system that could be scoped by something else and accidentally reach across.

The platform-wide operator role that can move between tenants for support is set in the deployment environment rather than stored as data, so no administrator inside the product can grant it to themselves.

A file you upload, or one synced from Google Drive, OneDrive, SharePoint or Dropbox, is cut into passages over its full text (not an excerpt, not the first few pages) and each passage is indexed by what it means as well as by the words in it. A search for "leave without pay" finds the clause headed "unpaid leave"; a search for the exact code on an invoice finds that too. Retrieval is instant; it is the reading and composing afterwards that take the time.

Every passage cites the file and its place in it, and resolves the same way a record id does. Each file is classified into one department when it arrives, and a department can search and read only its own department’s files: the same boundary that governs its connectors. The passage text is sent to Voyage AI to be embedded; nothing from your connected systems is. The security page says exactly what each processor receives.

Yes, when it holds a series of figures, such as revenue by month or headcount by department. The chart is drawn only from the figures in that answer, and a projected figure is drawn differently from a recorded one.

Each point opens to the record it came from, the same as clicking a figure. A chart costs no extra credit.

Yes. Ask about the future, such as what cash will look like at year end if you hire five people, and the answer works the forecast out from your own figures: what is in the bank, what a hire costs, how fast customers pay. Each of those opens to the record it came from, and a forecast always starts from at least one of them.

Every assumption is written beside the number it changes, with what it rests on, such as customers paying in 54 days or five people starting on 1 November. An uncertain assumption gets a low and a high, and the answer shows the range. Change an assumption on the answer and the forecast is worked out again at once. That uses no credits.

On Business and Enterprise, an assumption can also come from your market, such as how fast it is growing, found by the Research department and cited by page and date.

Because nothing is precomputed. Each department reads your live systems at the moment you ask, and the citations are verified before display. Speed would require caching, and a cached answer about a business that changed yesterday is a confident wrong answer.

It shows both figures, states which one it used, and explains why the other reads differently. For example, the CRM shows $14.80m booked this year and the ledger $12.90m of revenue. Revenue is Finance’s word, so the answer uses $12.90m and shows the bridge: multi-year contracts, orders not yet shipped, credit notes for returns, renewals never logged as deals, and exchange rates.

No. There is no unified model to build, because the architecture assumes the parts of a business will never agree on one. Each department is read from its own systems directly. If you already have a warehouse, it can be three connected systems among several.

It cannot, by design. Every connection is read-only, credentials are scoped per department, and the permitted calls are declared and enforced by the runtime. An answering system that could also write is a different risk profile than most finance teams will accept.

It works, narrowly and honestly. Finance is the usual place to start. You get complete answers to finance questions and clearly labelled stand-in answers or refusals elsewhere, which tends to make the case for the next connection better than a slide would.

Operators and finance leaders in businesses complex enough that the answer to a simple question lives in several systems, and consequential enough that a wrong figure in a board meeting, or a plan built on one, is expensive. Typically 50 to 2,000 people, with at least three systems of record.

In credits: one per part of the business a question consults, plus two when the answer had to be composed. Readers are unlimited on every plan; you pay for who may spend. Starter is $99 a month, and every paid plan can start with a seven-day free trial, no credit card required, or be bought straight away.

Yes, on Business and Enterprise. Once an admin turns on the public web on the Company page, the Research department looks into your competitors, your market, trends in your industry, news about your customers, suppliers and rivals, and regulation that affects you. It starts from what it already knows about your business: your website, what your team has taught it, your house rules and the documents you file to Research.

It also answers alongside your other departments. Ask how your revenue growth compares with your main competitors, and Finance reads yours from your ledger while Research finds theirs, in one answer. Every claim from the web cites the page and its date. Your own figures always come from your own systems, and where the web does not settle something, the answer says so. A question that uses the public web costs one credit more.

Yes, when you say so. Business context is public information about your market, competitors and industry, read beside your own systems: news, published results and filings, regulation, industry reports, public rates and holidays. It is off until an admin turns it on from the Company page, and then a department may search the public web and read public pages beside your own systems, such as a competitor’s filing, an industry report, a new regulation, a public holiday or a headline. Business context belongs to the Research department, and what each source reaches is on the Product page, under Research.

What comes back is treated as what it is: public information, cited by the page and the date it was read, never a source for a figure of yours. A department is told to prefer the primary source and to read the page rather than trust a search snippet. Every use is written down, and a question that draws on Business context costs one credit more than it otherwise would. Available from Growth, where other departments use it for their own figures; on Business and Enterprise the Research department also answers from it.

Yes. As you use it, each department learns your company: how it counts things, where its figures live, which queries and reports worked, which did not, and how you like answers set out. When somebody corrects an answer, that is learned too, and so is what the documents you file into each department show about how you work. It applies to the next question automatically. A note drawn from a document leaves when that document is deleted or moved, and nothing is learned from your website’s pages.

What is kept is short notes of that kind, for your company only. Never figures, names or records: those are always read fresh from your own systems and cited. An admin can switch learning off on the Company page. Notes are deleted with your company. Learning is on by default, on every plan.

No. Records are read to compose your answer and are not copied into a training set or a shared warehouse. What a department learns about how your company works is kept as short notes for your company alone, and is not used to train any model. Each answer carries the record ids behind its figures, so it stays auditable wherever it is kept.