30-minute quick start

From sign-in to your first cited answer. If you can spend half an hour, you'll know whether Knowledge fits your team.

This is the path our most-successful customers take in their first hour. Six steps, roughly five minutes each. Skip anything that doesn't apply to your job.

1. Sign in (1 min)

Go to ask.olyteck.com and click Sign in with Microsoft. Knowledge uses Microsoft Entra ID (Azure AD) — the same account you use for Outlook and SharePoint. The first person from your organisation to sign in becomes the workspace admin; everyone else is added as a member (see Invite members & roles for how to flip that).

2. Create your first project (2 min)

A project is a private folder of documents. Searches, agents and add-in calls can all be scoped to one project at a time, so different teams don't see each other's documents by accident.

In the sidebar click Projects → + New project. Pick a name that describes the collection, not the file type. Good names: Bid library, Trust & security, Engineering — past designs, Acme Corp account. Bad names: PDFs, Stuff, 2024.

3. Upload one real document (3 min + processing)

Click Upload in the sidebar, drop a file in, pick the project you just made. Supported formats: PDF, DOCX, XLSX, PPTX, CSV, TXT, MD, plus PNG/JPG via OCR. Up to 64 MB per file by default.

The Jobs panel shows live progress: queued → extracting → embedding → completed. A 30-page PDF usually completes in 15-30 seconds. A 200-page scanned PDF (OCR) might take a couple of minutes.

See Projects & uploads for the full list of supported formats and how OCR works.

4. Ask your first question (1 min)

In the sidebar click Search. Type a question that you genuinely don't remember the answer to (more interesting than a softball). Hit Ask.

Knowledge returns a synthesised answer with numbered citations — click any [1] to open the source drawer and see the exact passage in the original document. This is the moment that earns trust over a generic chatbot: every claim points back to a paragraph you can verify.

See Search, chat & citations for retrieval modes and follow-up questions in a thread.

5. Save your first agent (5 min)

An agent is a saved question template: prompt, model, scope, output shape. You build it once and call it from the browser, from Word, from Excel, or from your own code.

Open Agents → + New agent and fill in three things:

• Name & description — what this agent does.
• System prompt — how it should behave. For a security questionnaire responder, something like "Answer ONLY from the provided control evidence. If the question asks about a control that isn't in the evidence, say so explicitly. Never claim coverage that isn't documented."
• Projects — tick which projects this agent can search.

Leave model, top-K, temperature, output mode at their defaults to start. Every field is documented here when you want to dial things in.

6. Pick a workflow surface (5 min)

Knowledge has four ways to call an agent on real work. Start with whichever one matches the job you have today — you can layer the rest in later.

• Browser — Doc review. Upload a Word or PDF document, fact-check it against an agent, triage the findings as accept / reject / note. No install. Best for PDFs (the Word add-in is .docx-only).
• Browser — Questionnaires. Upload a security questionnaire / RFP / DDQ .xlsx, pick the question column visually, download the filled file with the original formatting preserved. No install. 100-row cap.
• Word add-in. Inline drafting, summarisation, fact-checking while you edit. Install → use.
• Excel add-in. Same questionnaire flow as the browser one, but runs in-place in your open workbook with no row cap. Install → use.