Context API - Paragon Documentation

The Context API lets your agent find relevant content from your users’ connected knowledge sources, then read more of the surrounding text when a result looks promising. Requests are sent to the Agent OS base URL and are authenticated with a Paragon User Token.

How it works

Create a Setup Link

Your backend calls Create Setup Link for a Connected User and gets back a hosted URL. Send that URL to your user.

Your user connects their apps

The Setup Link opens a hosted page where your user connects the apps they want your agent to search, such as Google Drive, Box, or SharePoint. Apps they’ve already connected show a green check.

The Setup Link app catalog, showing connected apps and available integrations grouped by category

Your user chooses what to sync

After connecting an app, your user picks the folders and files to include in their searchable content. The page shows how much content has been discovered and is searchable, and the sync status of each selection.

The Manage synced content page, showing discovery progress and a list of folders with checkboxes and sync status

Search the content

Once content is synced and indexed, your agent finds relevant passages with Search Context and reads more of the surrounding text with Read Segment Context.

Endpoints

Endpoint	Use it to
Create Setup Link	Create a hosted link where a user connects the sources to sync.
Search Context	Find the passages that best match a query.
Read Segment Context	Read the text before and after a passage returned by search.

How search results are organized

Search returns one result per record. A record is one item (for example, a file or page) in one sync. If the same item is included in two syncs, it appears as two results. Each result includes its best matching passages. Every passage has a pointer that you can pass to Read Segment Context to retrieve more of the surrounding text.

Supported sources

You can limit a search to specific sources with the sources field: onedrive, sharepoint, dropbox, box, confluence, googledrive

Respecting user permissions

Search results only include content the user is allowed to access in the source app. Use identity to tell Agent OS which account to apply on each source.

Field	Where	Description
`identity`	Request body on search. Query parameter on segment read.	The account whose access applies. Pass a string to use one account on every source. On search, you can instead pass an array of 1 to 20 `{ principal, source }` entries to use a different account for each source. The segment read endpoint only accepts the string form.
`email`	Request body or query parameter	Deprecated. Use `identity` instead.

The principal can be an email address or a provider-specific user ID. When you use a provider-specific ID rather than an email, you must also set source.

{
  "query": "q2 roadmap",
  "identity": [
    { "principal": "[email protected]", "source": "googledrive" },
    { "principal": "00uabc", "source": "sharepoint" }
  ]
}

​How it works

​Endpoints

​How search results are organized

​Supported sources

​Respecting user permissions

How it works

Endpoints

How search results are organized

Supported sources

Respecting user permissions