Data sources and knowledge
Business-Gym grounds its department heads on your documents. Two routes get them in: a Microsoft 365 sync over the OneDrive or SharePoint folders you pick, and direct upload on the Knowledge Base page. Both are one-way — text flows into the knowledge index, and nothing is written back into the source.
This page covers what those two routes do today, how a document reaches an answer, and which parts are not built yet.
Which sources can Business-Gym read today?
Section titled “Which sources can Business-Gym read today?”Two, and they are both on the Knowledge Base page, listed under Build in the sidebar. Its + New menu offers Add URL, Add files, Create text, Create folder and Sync Microsoft 365 — that menu is the complete set of ways knowledge gets in.
Microsoft 365 sync connects OneDrive or a SharePoint site library, lets you browse and multi-select folders, and re-reads them on a sync run. Direct upload takes PDF, DOCX, XLSX, CSV, TXT and MD files up to 10 MB each, or a whole folder whose subfolder structure is recreated inside the Knowledge Base. You can also paste a single URL or type a text document.
Confluence is not available yet. It is catalogued in the product as a known future source, but there is no connector to configure and nothing to try.
How does a document reach a head’s answer?
Section titled “How does a document reach a head’s answer?”Your source text is split into chunks, embedded, and stored in a per-tenant index. When a head answers, it retrieves the chunks that match the question within its own scope and drafts from those.
Both live routes point one way into the knowledge index — a sync reads a source and never writes to it. Confluence is drawn dashed because it is catalogued but not built. The draft at the end still goes to a person.
The draft at the end is the point. Knowledge makes a proposal accurate; it does not make it automatic. See the inbound → approve → execute loop for what happens after the draft.
Which head can read a document?
Section titled “Which head can read a document?”Only the one you routed it to. Every synced folder and every upload carries a scope: Tenant-wide,
visible to every head, or a single head, which stays in that head. The scope selector is on the upload
modal and on each folder you pick in the sync flow.
Retrieval enforces it. A head reads its own scope; the CEO-Planner reads tenant-wide. A document routed to one department is not reachable from another.
Before a sync saves, a “Review before you sync” dialog groups your picked folders by destination so you can see what is about to become tenant-wide.
Does connecting SharePoint change anything in SharePoint?
Section titled “Does connecting SharePoint change anything in SharePoint?”Indexing never writes back — that is enforced in the product, not a setting you can flip. The sync reads the folders you selected, extracts their text, and leaves the files where they are. It also does not reorganise your SharePoint or change any sharing permission.
The connect flow does offer an access mode toggle, and it defaults to Read-only. Read-write is a separate, deliberate grant for approval-gated writes by a head — it is not what the knowledge sync uses, and choosing Read-only loses you nothing on this page.
What happens when a connection expires?
Section titled “What happens when a connection expires?”Microsoft access tokens are time-limited, so a working connection eventually needs renewing. Within fourteen days of expiry, the connection card shows an amber “Token expires in Nd” pill next to a Renew now button that restarts the Microsoft consent flow.
If it has already lapsed, the card switches to “Reconnect needed” with a Reconnect now button. Neither one loses your picked folders or your synced documents.
How do I know a sync actually worked?
Section titled “How do I know a sync actually worked?”The connection card reports the real run, not a reassurance. It shows the folder count, the time of the last sync, the last run’s counts, and — when files failed — a View N issues drill-down that groups failures and separates temporary problems from permanent ones. Temporary ones have a Retry failed button.
The upload path counts the same way. After a batch you get a tally of what was indexed, what was unchanged and what was skipped — a file that is too large, has an unsupported type, or yields no text lands in skipped, never in indexed.
Above the document list, a coverage meter shows how much each head has to work with as a coarse level — none, thin, moderate or good — with a “needs more data” nudge where a head is empty. It deliberately does not show a percentage, because a made-up number would read as precision that is not there.
What is not available yet?
Section titled “What is not available yet?”Confluence, Company Profile distillation and freshness reporting are all absent today, and the two routes above are the whole picture. If a plan of yours depends on one of these, it depends on work that has not shipped.
- Confluence — catalogued as a future source, no connector to configure.
- Company Profile distillation — turning your corpus into a structured company profile you can review exists in the codebase, but there is no screen or control for it.
- Freshness reporting — nothing flags an indexed document as gone stale. The last-synced time on a synced folder is what you have.
None of these change the floor: heads propose, you approve, and external sends and money stay human-approved. Read autonomy and the hard floor for why that does not move, or working with a head for what a head does with the knowledge you give it. For the read-only analytics connectors — ads, CRM, website behaviour — see Integrations.
Who reviews the Company Profile my heads will rely on?
Section titled “Who reviews the Company Profile my heads will rely on?”Nobody, today — and nothing produces one to review either. The rules for that review already exist in the codebase, but no part of the product distils a Company Profile from your documents, and there is no screen on which a person could walk one through.
The intended shape is clear enough to plan around. A Company Profile is a short, structured description of your company — company facts, the founder, affiliated companies, products, services, history, key roles and key customers — distilled from the documents you have already added. It moves through three states: draft, in review, and published. It cannot go from draft straight to live; it has to pass through a person, and the approver is recorded on the profile.
Two accuracy rules travel with it. Every statement has to cite the document it came from: a statement that cannot be tied back to a document actually in your knowledge base is dropped as the draft is assembled, and a draft still carrying an uncited statement can be neither submitted for review nor published. And until a profile is published, your heads read nothing from it — a draft or in-review profile is simply empty as far as a head is concerned, so an unapproved description of your company can never shape an answer.
So what grounds your heads today is the documents themselves, through the two routes above: the Microsoft 365 sync and direct upload. Who at your company would hold the reviewer role, and when that screen arrives, are not answered anywhere in the product yet.