What happens to items after 48 hours?

Feed items and the raw scrapes behind them are hard-deleted by a job that runs every fifteen minutes. They are not archived or hidden; they are gone.

What if I need something from last week?

Save it when you see it. Saved items are permanent, and the image is copied into Prowler's own storage so it survives even if the original post is deleted.

Does the 48-hour window affect Trending?

Trending is computed over a seven-day window of scores, but the underlying items still expire at 48 hours. In practice items older than two days have decayed to near zero weight anyway, because the score halves every 24 hours.

Isn't deleting data a loss of value?

For an archive product, yes. For a monitoring product, an unread backlog is negative value: it hides today's signal behind yesterday's. The save list captures the value that was actually there.

Why your feed should forget: the case for a 48-hour monitoring window

Prowler deletes every feed item 48 hours after it was collected. Only items you explicitly saved survive. This is the design decision people ask about most, and it is deliberate: a monitoring feed that remembers everything turns into an archive within a month, and archives are where signal goes to be ignored. This post explains the reasoning, the mechanics, and what it means for how you use the product.

The failure mode of monitoring tools

Every monitoring tool starts as a stream and ends as a backlog. The stream is useful: you open it, the top items are what matters now, you act. The backlog is the same tool three weeks later: 1,400 unread items, a scroll bar the size of a pixel, and a growing suspicion that the important thing is somewhere in the middle. At that point most people stop opening it, which is the worst outcome a monitoring tool can have.

RSS readers made this failure famous, but keyword listening suites do it too; they just call the backlog a "mention history".

The cause is not volume. It is that the tool has two jobs, alerting and archiving, and optimises for the second because storage is cheap and deletion feels like loss.

A radar has one job

Prowler is a radar. The screen should show what is on it now. Two days is the window in which a post is still news, still reactable, still worth a reporter's or a marketer's attention. Past that, a post is either something you already acted on, or history.

So the rule is simple: feed items and the raw scrapes behind them are hard-deleted after 48 hours, by a job that runs every fifteen minutes. Not archived, not hidden behind a filter. Gone.

Three things follow:

The feed is always readable. Whatever your source count and cadence, the feed holds at most two days of ranked items. You can read it top to bottom.
Ranking stays honest. Trending halves an item's weight every 24 hours. Combined with the window, a post that was huge on Monday cannot sit at the top of Thursday's feed.
Attention is the scarce resource, and the product treats it that way. Nothing competes with today for your attention except today.

Saving is the other half

Deletion only works because saving is permanent and cheap. When you save an item:

It leaves the 48-hour window and is kept until you delete it.
Its image is copied into Prowler's own storage. Social platforms serve images through signed links that expire within hours or days, so a saved item that merely bookmarked the original URL would show a broken image within the week. Copying is the only way a save can outlive the day.
It is deduplicated by content: saving the same post twice, or the same image from two platforms, keeps one copy.

The result is a Saved list that is entirely things you chose, in your own storage, that will still look right in a year. That list is your archive. It is small because you curated it, which is what makes it useful.

What we do not keep, and why

Feed thumbnails are not copied. They live for 48 hours, comfortably inside the platforms' link lifetime, and copying every thumbnail for every scan would be a large amount of churn for nothing.
Raw scrapes are deleted with their items. They exist so every summary can be traced to its source while the item is live. After that, keeping a copy of other people's posts serves no one.
Source logos are the exception. They are copied once and refreshed only when the bytes change, because a source outlives any single post.

How to work with a window

The window changes habits slightly and, we think, for the better:

Read daily. Two days of buffer means a missed day is fine and a missed week is not. That is roughly the right pressure.
Save while reading. If a post might matter next week, save it now. Saving is one click; regretting is not.
Use Important and groups for focus, not for memory. Starring a source filters today's feed. It does not extend anything's life.
Trust Trending for "what has been building". Its seven-day scoring window looks back further than the feed does, but the decay means it is still about the present.

In Prowler: the retention job runs every fifteen minutes; Saved News is the permanent list; each saved item stores its image in Prowler's own bucket, re-uploaded only when the content hash changes. The feed shows a "New in 24h" count so you can see at a glance how much of the window is fresh.

The principle

Monitoring tools compete on how much they can collect. The useful metric is how little you have to read to know what changed. Deleting on a schedule is the most direct way to optimise for the second, and saving is how you keep the exceptions.

Related: what an agentic radar is, two agents, one clean feed. Or try the window for yourself with three sources.