RSS

Add RSS and Atom feeds, import OPML subscriptions, and understand how feed entries appear in the graph.

AgentGraph can index RSS and Atom feeds without API credentials. Each feed becomes a Folder entity, and each article or entry in the feed becomes a Document entity linked back to that feed.

RSS feed URLs are stored as connector configuration in ~/.agentgraph/config.toml by default, or ~/.agentgraph/config.yaml if that file exists. They are not stored in credentials.json.

#Install

Install every first-party connector:

uv sync --extra all

Or install only RSS support:

uv sync --extra rss

#Add feeds

Add one or more feed URLs:

agentgraph connector rss add https://example.com/feed.xml

AgentGraph validates every supplied URL as an RSS or Atom feed before saving it. HTML pages and invalid feeds are rejected without changing the connector configuration. A successful add queues an RSS poll immediately.

#Remove feeds

Remove one or more exact configured feed URLs:

agentgraph connector rss remove https://example.com/feed.xml

Removing a feed also deletes its feed Folder and relationships. Indexed articles remain in the graph and expire under the normal retention policy.

#Import OPML

Import an OPML export from a feed reader:

agentgraph connector rss import-opml feeds.opml --all

To choose specific feeds:

agentgraph connector rss import-opml feeds.opml --select 1,3-5

In an interactive terminal, omit --all and --select to choose feeds with a checkbox prompt.

#Sync feeds

Refresh every configured RSS feed immediately:

agentgraph poll rss

The RSS connector also polls configured feeds in the background when agentgraph serve is running.

#Browser observations

After indexing, a small set of article URL-prefix patterns derived from known entry links are exposed to the browser extension. The extension refreshes those patterns periodically and can then report observation for matching article pages.

The prefixes are only an extension-side eligibility filter. The server attributes observations to RSS only when the observed, normalized URL exactly matches a previously indexed RSS entry. Unknown pages under a matching article prefix are ignored.

RSS feed Folders are durable configured sources and are not browser-observable. Observing an article URL updates only the existing RSS Document whose metadata.web_url exactly matches the normalized URL. Polling and hydration do not count as observations or extend observation-based retention. See Entity retention.

#Query articles

RSS entries are indexed as Document entities with platform=rss.

agentgraph query --type Document --filter platform=rss --limit 20
agentgraph search "release notes" --platform rss --limit 10

Each article document includes the entry title, summary or content, source link, author metadata when available, and published or updated timestamps when the feed provides them.

#Article authors

Feed authors — Atom <author>, RSS <author>, and <dc:creator> — become Person entities. Every author credited by the feed or one of its entries is linked to the feed Folder, and authors are linked to their articles, by authored edges. Entries that declare no author of their own inherit the feed-level author, per the Atom spec. Authors are identified by email address when the feed supplies one and by name otherwise, so a name-only author is shared across every feed that credits that name.

agentgraph search "Matt Ridley" --type Person
agentgraph edges <person-id>