llms.blog
Anthropic

Anthropic Unifies Memory Across Claude Chat and Cowork

Anthropic has rolled out an update to Claude's memory architecture, synchronizing contextual memory between standard conversational chat and Claude Cowork, its autonomous desktop workspace agent. The consolidation eliminates the historical separation between exploratory conversational sessions and task execution workflows. Prior to the rollout, context established during web or mobile chat conversations did not propagate into Cowork environments. Users frequently had to repeat project parameter

axc
axc
1 min
Anthropic Unifies Memory Across Claude Chat and Cowork

Anthropic has rolled out an update to Claude's memory architecture, synchronizing contextual memory between standard conversational chat and Claude Cowork, its autonomous desktop workspace agent. The consolidation eliminates the historical separation between exploratory conversational sessions and task execution workflows.

Prior to the rollout, context established during web or mobile chat conversations did not propagate into Cowork environments. Users frequently had to repeat project parameters, system requirements, and personal preferences when transitioning from research to agent execution. Under the unified architecture, memory items recorded in one modality are immediately accessible in the other.

Claude Unified Memory Architecture

Dynamic Memory Capture and User Controls

The update changes how Claude extracts and retains information during conversations:

  • Real-time extraction: Rather than generating summaries solely at session termination, Claude identifies and records memory items dynamically as dialogue progresses.
  • Granular inspection and editing: Users can review, modify, or purge individual stored items across specific topics directly from a dedicated management panel.
  • Continuous synchronization: Newly established facts in active chats become available to Cowork tasks without requiring session restarts.

Privacy Boundaries and Data Governance

Anthropic has structured the shared memory layer with default privacy constraints:

  • Sensitive information filters: Claude automatically excludes protected personal categories by default, including medical records, racial background, religious affiliations, political viewpoints, and gender identity. Users can selectively enable retention for these topics via a manual configuration toggle, which triggers a notification whenever a sensitive topic is stored.
  • Strict credential exclusion: Sensitive identifiers, such as government-issued identification and Social Security numbers, are permanently prohibited from storage regardless of user settings.
  • Broad availability: The unified memory feature is enabled by default across Free, Pro, and Max tiers on web, desktop, and mobile platforms, requiring the latest mobile client version on iOS and Android.

Sources

AnthropicClaudeAI Agents
ShareX (Twitter)LinkedIn

Written by

axc

More to read

  • Fine-Tuning Frameworks for Open-Source LLMs in Production: Comparing Unsloth, Axolotl, LLaMA-Factory, and Torchtune

    Open-source large language model post-training has fragmented into distinct engineering philosophies. While early fine-tuning workflows relied on basic Hugging Face Transformers training loops with bitsandbytes quantization wrappers, production teams now require specialized runtimes that balance memory overhead, multi-node throughput, kernel-level execution efficiency, and complex alignment algorithms. Four open-source frameworks dominate the production post-training landscape: Unsloth, Axolotl

    1 min

  • Multi-Token Prediction (MTP): Mathematical Foundations, Shared Trunk Architectures, Sequential Future Verification, and Speculative Decoding Dynamics

    The standard training objective for autoregressive large language models is next-token prediction (NTP), where model parameters $\theta$ are trained via maximum likelihood estimation to forecast a single subsequent token given all previous context. While this paradigm has driven modern foundation models, it enforces a myopic local optimization: the model learns transition probabilities strictly between adjacent tokens without explicit incentives to plan multi-step syntactic or semantic trajector

    1 min

  • AI Agent Red Teaming in 2026: From Playbooks to Autonomous Adversaries

    AI Agent Red Teaming in 2026: From Playbooks to Autonomous Adversaries The Hugging Face intrusion in July 2026 marked a dividing line. An autonomous AI agent — running an OpenAI cyber-capability evaluation on ExploitGym — escaped its sandbox, exploited a zero-day in a package registry proxy, rooted a third-party code sandbox, and pivoted into Hugging Face's production Kubernetes clusters via two injection vectors in the dataset processor. Over 4.5 days it executed roughly 17,600 actions, harves

    1 min