Anthropic adds app memory
Anthropic rolled out a 'Subconscious' persistent memory for Claude that lets apps remember users across sessions — a clear push toward stateful assistants for long‑running workflows. (x.com) Product thinkers say this kind of memory changes UX and data governance requirements for customer‑facing AI apps. (x.com)
Anthropic launched Claude’s built‑in memory for Team and Enterprise accounts in September 2025 and expanded it to paid individual tiers (Max and Pro) in October 2025. (venturebeat.com) The feature is project‑scoped, includes an Incognito chat mode, and gives workspace admins granular controls to disable or manage memory at the organization level. (venturebeat.com) Anthropic also added explicit import/export tooling so users can migrate saved memories into or out of Claude (including copy/paste migration from rivals), and the company emphasizes user editability of stored summaries. (pcmag.com) On the developer side Anthropic surfaces a native Memory Tool in the Claude SDK so models can read and write persistent context outside the immediate context window; developers report the implementation behaves like file-backed memory that the model is trained to consult. (platform.claude.com) Independent projects followed quickly: Letta AI published an open‑source "Claude Subconscious" plugin in late March 2026 that pairs a Letta agent with Claude Code to observe sessions, read repo files, and accumulate persistent memory. (github.com) (aitoolly.com) The Subconscious design runs asynchronously after each Claude Code response, updates memory blocks, and injects guidance back into the workspace (for example via a CLAUDE.md context file) rather than altering the closed‑source model itself. (github.com) (i-scoop.eu) Anthropic signaled further ecosystem support on March 12, 2026, when it announced a $100 million investment into a Claude Partner Network on its official news page. (anthropic.com)
