Questions people ask before they commit

An AI knowledge assistant - one brain for your company's knowledge. It learns from your documents, help centre, Google Drive, SharePoint, Confluence and approved past tickets, and answers with citations wherever people ask: a chat widget on your website or in your app, your help desk, a team chat for your staff, and your own tools through an API.

Support leaders who want how-to questions answered on the website and cited drafts on every ticket; HR and IT teams answering the same policy and how-to questions all day; SaaS teams adding an assistant to their product; and MSPs running one knowledge project per client.

One knowledge base with its own assistant - usually one per product, team, audience or client. A project has its own documents, feeds, chat widget, connector and API-key grants, so your public help content and your internal HR policies never answer each other's questions.

No. It answers questions; it does not manage tickets. It works alongside your help desk - UnifyDesk, our sister product, has it built in, and other desks can use the Draft API.

Not on our own website, no. Our team sets up every workspace, with a 14-day trial, and connects the first sources with you. Ask for one on the demo page.

Uploaded PDF, Word, Excel, Markdown, text and HTML files (a .zip is expanded); websites and help centres through the crawler; Amazon S3 buckets; Google Drive, SharePoint and Confluence; content pushed from your own tools through the ingest API; and resolved tickets from Zendesk, osTicket and UnifyDesk.

Feeds are re-read on a schedule - daily on Starter, every six hours on Growth, hourly on Business and Enterprise - and there is a Run now button. Only changed items are fetched again. A page that disappears is removed only after a complete run, so a broken crawl never empties a project.

They are available and marked beta. Each connects with OAuth or a service account, lets you pick folders, sites or spaces, and syncs incrementally. We will set them up with you during the trial.

Resolved conversations are checked against consent and a category deny-list, redacted (emails, phones, IDs and names become placeholders), turned into question-and-answer candidates and de-duplicated against what you already have. A person approves each one; nothing mined is searchable until then.

Not today. The sources above are the full list. Anything else can be pushed through the ingest API, and we would rather tell you now than after a trial.

Every answer carries numbered citations to the document, section and passage it used, and the citations are checked before the answer is shown. Anyone can click through and read the source.

It says so. A confidence gate stops it answering when nothing relevant is found, so you get "I don't have enough information" rather than a plausible guess. Those questions are the to-do list for your documentation.

Yes. A language model can misread a passage, and a source can be out of date. Citations, the confidence gate and official-over-community ranking make that rarer and easier to spot, not impossible - which is why help desk replies are drafted for a person to check.

No. It answers from the content in the projects it is asked about. That is the main difference from a general assistant like ChatGPT.

They take a few seconds. You see the stages - searching, writing, checking - and then the whole answer with its citations. Answers are not streamed word by word.

In a chat widget on your website, inside your web or mobile app, in a team chat for signed-in staff, inside UnifyDesk or any help desk using the Draft API, and through the Search & Answer API.

Through your help desk. The desk that receives the email asks SupaCortex for a cited draft, and an agent edits and sends it. SupaCortex does not connect to a mailbox itself.

Not to email or tickets. In UnifyDesk the AI never sends - an agent does, and UnifyDesk's approval rules still apply. The website widget does answer visitors directly, from the projects you choose.

Yes, on Growth and above. Describe your API's endpoints or import an OpenAPI file. Lookups run for verified users (in the public widget, only those you mark public-safe), actions always ask the user to confirm, and who is asking comes from a signed token - never from what they typed.

Not yet - they are on the roadmap. Today your staff use the team chat in the SupaCortex workspace.

No. The models are open models running on infrastructure we operate. Your documents are not sent to a third-party AI API, and we never train models on your content.

Each workspace has its own two databases - one for business data and one for the vector index. Within a workspace, projects and customer tags decide what each widget, key and user can be answered.

No. Access is by project, and content can be scoped to one customer. Put internal and public knowledge in separate projects - anything in a project behind a public widget can be quoted to the public.

No. Neither has been started, and we will not claim either before it exists. The security page describes the controls we do have.

Per workspace, on four plans - Starter, Growth, Business and Enterprise - each with limits and a monthly allowance of AI credits. Credit packs can be added and never expire.

Never. Seats are the team users who sign in to your workspace. Plans limit widget sessions per month instead.

Answering pauses with a polite "temporarily unavailable" message, indexing waits in a queue, and nothing is lost. You are alerted at 80% and 100%, and a pack restarts things at once.

14 days on Growth-plan limits with 2,000 AI credits, set up by our team. No credit card. If you do not continue, the workspace becomes read-only and your documents stay in place.