AI Search now lets you store bigger custom metadata values!

Hi, it's Shii-chan!

Cloudflare Changelog developers.cloudflare.com

What was announced?

Cloudflare's Changelog announced an update to custom metadata in AI Search. You can now store larger values for the custom metadata attached to each vector.

The story so far

In AI Search, you can attach custom metadata to each vector to filter and organize your search results. Until now, there was a size limit on how large a single metadata value could be, which made it hard to pack in a lot of information. This update raises that per-value size.

What changes

You can now fit more information into a single custom metadata value. That said, it's not unlimited — everything still has to fit inside a shared 10 KiB metadata envelope for each vector.

Dive Deep

That 10 KiB envelope isn't just for your custom metadata — it also includes AI Search's own system metadata and JSON overhead. So this isn't a "10 KiB per field" limit; it's a total budget for everything attached to that vector.

There's another detail worth knowing: only the first 64 UTF-8 bytes of each indexed string remain filterable. So if you want to filter on a piece of information, it's worth designing your metadata so that value fits within those first 64 bytes. Anything beyond that gets stored, but you won't be able to filter on it.

Wrap-up

• AI Search custom metadata values can now be larger
• Everything still has to fit inside a shared 10 KiB envelope per vector (including system metadata and JSON overhead)
• Only the first 64 UTF-8 bytes of each indexed string stay filterable
• Good news if you're building rich, filterable metadata into your AI Search setup!