Company knowledge

Enterprise AI search that checks who's asking.

Ask a question in plain words and Klaie answers from your company's documents, connected apps and the web, with a citation for every claim. Before it opens a single file, it checks whether you're allowed to.

What did we agree with Acme on renewal pricing?
Read 2 documents and 1 email you have access to

You offered Acme 12% off a two-year renewal if they sign before 30 November1. Their finance lead asked for net-60 payment terms, and nobody has replied yet2.

1CompanyAcme renewal terms.docx2MailRe: Renewal, Acme
Example: someone asks Klaie “What did we agree with Acme on renewal pricing?”

Search everything you can open, and nothing you can't.

Access is checked in the database

Permissions are applied inside every search, not after it. A file you can't open is never retrieved, so no model reads it on your behalf.

Every claim has a source

Answers cite the document, email or web page they came from, and each citation opens the exact source.

One search across your tools

Uploaded files, Google Workspace, Microsoft 365 and the web are searched together, so you don't need to know where something lives.

Restricted, not guessed

When something is restricted, Klaie says it can't find anything you have access to and lets you ask for access. It doesn't fill the gap with a guess.

Why permissions come before the model

Most AI assistants retrieve whatever they can reach and rely on the model, or a line in a prompt, to keep confidential material out of the answer. That's a promise a language model can't reliably keep.

Klaie works the other way round. Retrieval runs as the person asking, with their organization, role and collection access applied in the query itself. When Jordan in engineering asks how the company's finances are doing, the restricted finance report is never fetched. Maya, who can open Finance, gets the numbers, and opening the confidential report is written to the audit log.

What Klaie can search

Everything is searched with the access of the person asking. Company sources and web sources are labelled differently, so you can always tell which is which.

  • Files your team adds: PDF, Word, PowerPoint, Excel, CSV, text, Markdown, HTML, JSON and images
  • Gmail, Google Drive (including shared drives) and Google Calendar
  • Outlook, Teams chats, OneDrive, SharePoint and Outlook Calendar
  • The web, through Klaie's own search and page reader

Answers you can check

Each answer carries numbered citations. Click one and the exact document, message or page opens, so checking a claim takes seconds. When sources disagree or don't cover the question, Klaie says so instead of smoothing it over.

Questions people ask

How is this different from a chatbot connected to our shared drive?

A typical connector indexes everything with one service account and then filters what the answer shows. Klaie applies each person's access during the search itself, so restricted content is never read on their behalf in the first place.

What happens when someone asks about something they can't access?

Klaie tells them it can't find anything they have access to. Where a collection is restricted, they can ask for access. The restricted content itself never reaches a model.

Does Klaie use our documents to train AI models?

Klaie never uses your content to train its own models. Answers on the Klaie 1, Kai 1 and Mol 1 tiers come from third-party AI providers under their own terms, and Klaie flags any tier whose terms let the provider learn from requests. Sensitive mode never sends anything to an outside provider.

Which file types can Klaie read?

PDF, Word (.docx), PowerPoint (.pptx), Excel (.xlsx), CSV and TSV, plain text, Markdown, HTML, JSON, and PNG, JPEG, WebP and GIF images.

Start with one question.

Setup takes about a minute. Bring in files and apps whenever you're ready.

Create a workspace

Free during early access. No card needed.