What Knowledge can read — workspace files, attachments, Slack, Jira, Gmail and the public web — how each source is kept private and permission-checked, and how citations work.

On this page

Knowledge only knows what you let it read. Each conversation, and each expert, has an explicit set of sources.

Source types#

Source What it covers How to add it
Workspace files PDFs, Word documents, Markdown, text and images uploaded to the workspace. Files
Attachments Files and images added to one question. + Files in the composer
Slack Channels and threads the connection can see. Slack
Jira Issues, comments and statuses in the connected site. Jira
Gmail Messages in a connected mailbox, private to its owner. Gmail
Web Public web pages and PDFs, read when needed. Web toggle in the composer
Memory Facts and preferences you asked Ysra to keep. Memory

Repository sources for Knowledge are coming; they appear as Soon in the composer.

Choosing sources#

  • In the composer, toggle Files and Web and pick connected sources.
  • On an expert, set default sources; conversations with that expert start with them.
  • By default, a conversation uses the expert's sources plus the workspace's ready mailboxes.

Knowledge searches company context only when your question needs it. Ask a general question and it simply answers.

Privacy and permissions#

  • Only ready, connected sources you own or were given can be searched.
  • Each document keeps its access rules. A private message stays visible only to the people allowed to see it.
  • Mailboxes are owner-private. One person's connected Gmail isn't searchable by others.
  • Ysra does not mirror every permission in your organisation. It searches the specific sources selected for the conversation or expert.

How citations work#

  1. Every source Knowledge reads during a conversation gets a number.
  2. The answer may cite only those numbers.
  3. If an answer cites something that wasn't retrieved, it is rejected rather than shown to you.
  4. The Sources panel lists each source with provider, excerpt, link and the time it was read.

For the web, a search result is only a lead: the page must be fetched and read before it can be cited. Public PDFs are read as documents.

Files and attachments#

  • Uploaded files are parsed and indexed so their content can be searched — see Files.
  • Images you attach are analysed and bound to the question you asked with them.
  • If a file fails to process, retry it from the Files page.

Sources are data, not instructions#

Everything Knowledge reads — a Slack message, an email, a web page — is treated as information to reason about. Instructions inside a source are never followed.

YsraDocs