Anthropic details agent memory: 'dreaming' lets Claude organize memory offline
adnan_hashmi · x · 2026-11-08
Anthropic's engineering team breaks down Claude's agent memory architecture—transcripts, memory notes, memory representations, and injected context—and explains how a 'dreaming' mechanism, inspired by sleep-time compute from Letta and UC Berkeley, uses idle time to fix mistakes and consolidate knowledge, plus the evals used to measure each component's contribution.
Related event: Anthropic's Engineering Blog Details Agent Memory with Offline "Dreaming"(14 posts)→
More from coding & agent
- Leaked account: Google-style review bureaucracy stalls startup Reflection, PRs take weeks to merge — SonglinYang4 · 2026-11-08
- New Vercel Hire Builds a Rex-Powered TUI for Spinning Up Agent Sessions — JohnPhamous · 2026-11-08
- Claude Code's "Don't Ask Again" File Exposes Passwords and Keys in 1,122 Public GitHub Copies — jedisct1 · 2026-11-08
- OpenAI Buys Pierre to Give Codex a Home for Agent-Written Code — shashib · 2026-11-08
- New language 'blood' claims to beat Rust with effect systems and SQLite-stored code — sull · 2026-11-08
- AI says '8 more minutes' while the command has already run for 17 minutes — heypearlai · 2026-11-08