Tore

Features / Self-serve

Help center

Answers people can find themselves

Build it three ways: write it, upload what you already have, or pull it in from a URL. Then it is searchable, on your own domain, and used as the source when a reply cites where an answer came from.

Help center
Search for an answer…

What it does

  • Write articles directly, in collections, with a real editor
  • Upload what you already have, documents and PDFs are parsed and indexed, not just attached
  • Import from a URL, so an existing help page becomes an article
  • Internal-only notes live alongside public articles, and are never shown to customers
  • Search understands meaning rather than matching keywords, so 'can't log in' finds the password article
  • Every AI answer cites the article it came from, so a customer can check it
  • Your domain, your branding, no Tore logo

Without it

The answer exists. It is in a thread from March

You have written this answer eleven times

How to rotate an API key, explained slightly differently in eleven threads, each one a little worse than the last because it was typed at speed between two other tickets.

The documentation exists. Nobody finds it.

The customer searches “can't log in.” The article is called “Resetting your password.” Keyword search returns nothing, so they open a ticket, and you send them the link they could not find.

The help center that rotted

Thirty articles written in one heroic launch week two years ago. Half of them describe screens that no longer exist, and everyone on the team quietly stopped linking to it.

How it works

Step 01

Build it from what you already have

Write an article in the editor, upload a file, or point at a URL. An uploaded PDF, Word document, Markdown file or HTML page is parsed into text and indexed, not just attached as a download.

What that includes

  • Write an article in the editor
  • upload a file
  • point at a URL
Step 02

Search understands the question, not just the words

Content is chunked and embedded, then searched two ways at once: meaning-based similarity and Postgres full-text search, with the two result sets fused. That is why “can't log in” reaches the password article.

What they see

t log in

Step 03

One article, one collection

An article lives in exactly one collection, so moving it means something and there is never a question about which copy is the real one. Collections are what customers browse when they are not searching.

Step 04

Internal notes sit alongside and never leave

A note is indexed for your team and for the AI that drafts replies, and is never served on the public help center. The runbook you would never publish still makes the answers better.

Step 05

Publishing is its own permission

Writing an article and publishing one are separate rights. Someone can draft all week without anything reaching customers until a person who is allowed to publish says so.

What it means for you

The answer you wrote once gets used a hundred times

Not just by customers reading it. The same articles are what the AI cites when it drafts a reply, so improving one article improves every future answer that touches it.

People solve it before they write to you

The tickets that never arrive are the cheapest ones you will ever handle, and the customer got their answer at 11pm without waiting for your morning.

An AI answer a customer can check

Replies cite the article they came from. A customer who does not believe the answer can open the source, which is the only version of AI support a skeptical person will accept.

The honest limit: Two pieces are not switched on yet. Importing a whole Markdown documentation repository is designed but not built, so today the on-ramps are writing, uploading and importing from a URL. The ask-a-question box on the public help center is also still showing a placeholder answer while it is wired to the live search behind it.