How many assets can BETTER! handle?

BETTER! is built for libraries of up to 50 million assets. In stress runs on a corpus carrying millions of metadata rows, search indexing ran flat at hundreds of documents per second with zero errors, per-asset metadata reads came back in milliseconds, and the ingest pipeline processed roughly 10,000 queued assets a minute. The measured rates stayed flat throughout — performance did not degrade as the corpus grew.

Does BETTER! slow down as the library grows?

The stress runs specifically measured for degradation: the search reindex rate stayed flat from the first document to the last, and metadata reads stayed index-backed and in the milliseconds with millions of metadata rows in place. Flat rates are what make a 50-million-asset design target credible — performance stays predictable, so maintenance windows are plannable instead of guesswork.

How fast can BETTER!'s AI agents work through a large backlog?

The ingest pipeline sustained roughly 10,000 assets a minute in stress runs. Deep AI description and transcription proceed continuously, paced by the AI models you choose (cloud or local). New uploads are processed on arrival, so large backlogs are typically a migration-day event, not daily life.

The Stress Runs

50 million assets. That's the design floor.

Every DAM says it scales. We publish the stopwatch. BETTER! is built for 50-million-asset libraries — and instead of asking you to take that on faith, we run a standing stress corpus — millions of metadata rows deep — through the operations that actually hurt at scale: full reindex, metadata reads under load, an agent backlog the size of an entire migration. Here's what the clock says.

The Numbers

Measured, not projected

50M

Assets — the design floor

The design target our certification runs are built around.

~330/s

Search indexing, flat, zero errors

Same rate at the first document and the last — no degradation curve in our runs.

<15ms

Metadata read, per asset

Index-backed, with millions of metadata rows in place.

~10K/min

Ingest pipeline throughput

Sustained, measured on the stress corpus.

<50ms

Search response

Keyword and semantic, typically well under 50ms in our runs.

The word that matters in all of this is flat. Plenty of systems post good numbers at ten thousand assets and quietly decay as the library grows. We measure for exactly that failure: in our runs, the indexing rate at the last document matched the first. Flat measured rates are the whole argument behind the 50-million design floor — they are what make performance planning possible at all.

Why This Matters

Agents are only useful if the floor holds

An agentic DAM multiplies the write volume of a traditional one — agents describing, enriching, transcribing, and indexing everything, all the time. That's precisely why we certify the foundations at scale: autonomy on top of a platform that degrades is just a faster way to hit the wall.

Migration-ready. Moving off a legacy DAM means arriving with your whole history on day one. The ingest pipeline drains at roughly 10,000 assets a minute in our stress runs; AI description and transcription then work through the library continuously, paced by the models you choose.
Predictable operations. A full reindex — the scariest routine operation a search-backed system has — ran at a known, flat rate with zero errors in our runs, which is what makes maintenance windows plannable rather than open-ended.
Honest methodology. These rates come from dedicated stress runs against a standing corpus built to punish worst cases, re-run after performance-relevant changes. When the numbers change, this page changes.

50 million assets. That's the design floor.

Measured, not projected

Agents are only useful if the floor holds

Bring your library. We're not worried.