Research · Sunday, November 8, 2026
Anthropic describes how Claude keeps memory
The File desk · Nov. 8, 2026, 6:41 a.m. ET
Confirmed1 primary sourcesAdded 6:41 a.m. ET, Nov. 8
Status
Confirmed that Anthropic published the engineering post on November 6, 2026. The description of agent memory and dreaming is Anthropic's account of its own system. This is not a new model announcement.
On November 6, 2026, Anthropic published an engineering post, "The components of agent memory." It says agent memory is knowledge Claude gains from experience and retains between sessions, saved as context it can read later, and not a change to the weights set in training. It says a process often called dreaming can organize that memory, correct mistakes, and create notes.1
How we know
Notes
- 1.Anthropic, "The components of agent memory," Published Nov 6, 2026, Team Engineering. "By 'agent memory,' we mean knowledge that Claude gains from experience—including interactions with users and tasks it's been assigned—and that it retains between sessions, saved as context it can read later. It excludes other ways models can learn or manage context, such as changes to their weights (the internal parameters set during training)." "This process is often called dreaming." "Claude can use dreaming to organize memory. This can correct mistakes, create notes." Retrieved 6:16 a.m. ET. https://www.anthropic.com/research/components-of-agent-memory
What is still unknown or disputed
- A measured score for the memory system. The post discusses evaluations. This brief does not extract a score.
- A new model name. The post does not announce one.
Primary sources
Every claim in this story is drawn from the documents below. If a fetch failed, that is recorded on the card.
Source 1
The components of agent memory
Anthropic · November 6, 2026
Anthropic describes agent memory as knowledge Claude retains between sessions, and dreaming as a way to organize it.
https://www.anthropic.com/research/components-of-agent-memory