Anthropic's Message Threads lets AI agents skip resending whole conversations
The public beta cuts wait times on very long requests by about a third, Anthropic says.
Anthropic said Tuesday that developers can now have Claude remember a conversation in progress, so their programs no longer resend the whole thing with every request. The San Francisco company released the feature, called Message Threads, as a public beta.
Normally, an app talking to Claude sends the full chat history each time, which gets slow as AI agents work through long tasks. With threads, the first request includes everything, and later ones send only new messages plus the ID of Claude's last reply.
Anthropic says threads are kept encrypted in memory, not saved to disk, and expire five minutes after the last request. It said that with threads on by default for most users of Claude Code, its coding tool, the typical wait for a first word on very long requests dropped about 35%. Token use and prices do not change.
The feature works on Anthropic's own platform and through Amazon Web Services, Microsoft Foundry and Google Cloud.
Source: Anthropic article and developer documentation, Nov. 3, 2026