Knowledge sources overviewМеню документации

Knowledge sources

Knowledge sources overview

What a connection is, how it differs from an integration, and how Harmona processes your data and keeps it in sync.

Документация доступна на английском и турецком, как и интерфейс Harmona. Эта страница показана на английском.

A connection is knowledge your agents read: uploaded files, a database, a document store, a mailbox or a set of web pages. You manage connections under Studio → Connections. Once a connection exists, you can attach it to an agent, a Worker or a workflow step, and they answer from it with citations.

Connections and integrations

Harmona has two ways to reach your data. They often work together, but they do different jobs.

ConnectionIntegration
What it isKnowledge Harmona reads, processes and keepsAn account an agent acts in, live
WhereStudio → ConnectionsStudio → Integrations
What agents doSearch, read and query itRun actions: search mail, create a ticket, post a message
FreshnessRe-synced on a scheduleLive on every call

Some sources need an integration first. A Google Drive, SharePoint, ReadMe, Zoho Desk or Outlook mail source reads through the account you connected under Integrations. See How integrations work.

Connection types

Click Create Connection on the Connections page and pick a type:

  • Internal Files: upload documents, spreadsheets and images. See Files and images.
  • Database: query PostgreSQL, MySQL, SQL Server, Oracle, Snowflake, SAP HANA or SQLite. See Databases.
  • SFTP Server, SharePoint, Google Drive, ReadMe.com Docs and Zoho Desk Tickets: sync a document store. These are documentation connections. See Documentation connections.
  • Web Crawler: fetch and index web pages. See Web pages.

Processed on connection, kept in sync

When you add a source, Harmona fetches its content and prepares it once, in the background:

  • Documents are converted to text and indexed, so agents can search them and cite the exact file.
  • CSV files, Excel sheets and Parquet files become SQL tables, so agents can count, sum and join across files.
  • Images are read by a vision model into a searchable description.
  • Optional deep analysis turns tables inside documents into data and builds an entity index.

After that, synced sources are checked again on a schedule. New and changed files are processed again, and files you removed at the source are dropped. A database is different: agents query it live, so there is nothing to copy.

Sync interval

One setting controls how often every syncing connection re-syncs. Go to Management → Organization Settings → General and set Sync interval under Connection Settings. You can enter a number of minutes, hours or days. The default is once a day.

To pick up a change right away, open the connection, select the source and click Sync Source. For a web crawl, use Sync Connection. The connection page shows Last Sync and a progress panel with completed, unchanged, deleted and failed items. Use Retry Failed Files when some files did not process.

Who can use a connection

Connections are shared like other Harmona assets. People see Can view, Can use or Can manage on a connection that was shared with them. If you can only view a connection, you see its content but cannot change it. Ask its owner for manage access. See Sharing and access.

Внимание

Deleting a connection cannot be undone. Its search index is deleted and it is removed from every agent and workflow that used it, so they stop answering from it.

Where to go next

Обновлено 2026-09-24