clauding.de
DE
• 2 min read AI-generated

Claude Developer Platform: New Beta Header Tidies Up Agent Memory

Copy article as Markdown

Anthropic introduced the 'agent-memory-2026-07-22' beta header. It makes listing agent memories more predictable — with stable ordering and clear rules for 'depth' and 'path_prefix'. Eight SDKs now send it automatically.

Anthropic Developer Platform Memory
Featured image for "Claude Developer Platform: New Beta Header Tidies Up Agent Memory"

No big model, no feature fireworks — but if you build with Managed Agents and their memory on the Claude Developer Platform, today is worth a closer look. Anthropic introduced the agent-memory-2026-07-22 beta header, noticeably tidying up how the memory endpoints behave.

What changes

The main target is listing memories (GET /v1/memory_stores/{id}/memories). Specifically:

  • Stable ordering: results now come back in a fixed, server-defined order. The order_by and order parameters are ignored.
  • Clearer depth: only 0, 1, or omitting the value are allowed. Anything else returns a 400 error.
  • More precise path_prefix: the prefix must end with / and now matches whole path segments instead of arbitrary substrings.

Important for migration: page cursors issued without the new header are not valid with it — you have to restart from the first page. On the memory store endpoints, agent-memory-2026-07-22 replaces the old managed-agents-2026-04-01; send both at once and you get a 400.

SDKs come along automatically

Eight official SDKs now send the new header on all memory store calls: Python 0.116.0, TypeScript 0.110.0, Go 1.56.0, Java 2.48.0, Ruby 1.55.0, PHP 0.36.0, C# 12.35.0, and the CLI 1.16.0. Update, and you get the new behavior without touching the header yourself.

My take

Updates like this are unspectacular, and that’s exactly the point: agent memory is growing up. Stable, predictable ordering and strict rules for depth and path_prefix sound like minor housekeeping — but they’re precisely the things that quietly break production agent systems otherwise. Anthropic is tightening the API so behavior stays reproducible. If you run Managed Agents, take the migration notes seriously, or you’ll be chasing unexpected 400s.

Sources: