Connecting sources
Point Miatz at the places your product truth already lives, and run them.
- Connect a documentation site, an API spec or a repository in one form
- Runs happen on a queue, so a large corpus never blocks your screen
- Re-running only processes what actually changed
Adding a source
Give it a name, choose the kind, and paste the address. Only ordinary web addresses (http and https) are crawled, and nothing is fetched until you press Run now.
Kinds we understand
Documentation sites and sitemaps, OpenAPI specs, GitHub and GitLab repositories, Confluence spaces, Jira projects, help centres (Zendesk, Intercom, Freshdesk), Notion workspaces, video transcripts, release notes, status pages, postmortems and support-ticket themes.
Secrets
This form stores no credentials, by design and by database constraint. If a connector needs authentication, it is configured through the sealed secret path rather than typed into a screen where it would end up in a configuration blob.
Running
Ingestion is queued work, not a page load: a corpus can be tens of thousands of pages. The queue retries with backoff on transient failures, and a re-run only reprocesses pages whose content hash changed.
Frequently asked
Who can connect or run a source?
Owners, org admins and enablement managers. Authors, reviewers and analysts can see what is connected but cannot change it — connecting a source decides what the whole academy is built from.
What happens when I press Run?
An ingestion job is queued. It crawls the source, mines the pages, hashes them for change detection, and re-scores the corpus. The screen shows when each source last completed.
Can I connect a private system that needs credentials?
Yes, but not through this form. The form deliberately refuses to store secrets, and the database rejects any connector configuration containing a credential in plain text. Private connectors are set up with your Miatz contact so the secret lives in the sealed secret store.
How often should a source run?
Whenever the underlying content moves. Most teams run on each release and let the scheduled runs catch everything in between — the point of the loop is that stale content is visible rather than silent.