Anthropic's Overnight Shocking Upgrade: Claude Gets "Permanent Memory", Changing the Game for Global Workers

Just now, Anthropic has once again set off a storm in Silicon Valley: Claude Cowork is reported to have been infused with "permanent memory"! From now on, AI will no longer be forgetful but will develop a permanent super brain. An AI office revolution is on the verge of breaking out.

Anthropic, which has stirred up the entire Silicon Valley, continues to drop new bombs.

Today, sources revealed that Anthropic is giving a major upgrade to Claude Cowork by injecting permanent memory into its knowledge base!

That is to say, from now on, Claude will no longer have a "goldfish memory." In its "permanent brain," it will start to automatically remember everything.

From now on, Claude will no longer be just a chatbot. This Cowork mode with permanent memory will completely revolutionize the AI office!

In addition, the Cowork mode will be merged with the Chat mode and become the default user interface of Claude Desktop.

Overall, the exposed internal instructions from Anthropic contain a wealth of information and also include other major upgrades.

Knowledge Bases

Claude Cowork becomes the main entry point

Unified UI + Artefacts sidebar

More powerful automated connectors (MCP)

Voice mode + Pixelate upgrade

Another well - known Silicon Valley leaker, Tibor Blaho, also revealed similar news.

Reddit users who have already used Cowork are excited and said: They can't wait! This feature will make collaborative programming much easier!

If all these features are successfully implemented, Claude will evolve into a true productivity partner - one that can collaborate long - term, execute tasks, and remember everything about you!

Someone in the industry revealed that their company is also developing a similar context engine internally. It is speculated that the ultimate goal of Anthropic's move is to reduce token consumption.

Sure enough, Anthropic is leading the way in Silicon Valley.

Knowledge base gives Claude "permanent memory"

According to the leak, Anthropic is working intensively on Claude Cowork's "permanent memory."

This "knowledge base" is a brand - new memory method designed by Anthropic for Claude.

The introduction of the knowledge base is the core new feature.

Even Claude Code doesn't have this feature, but Cowrok will be the first to have it.

Internal instructions of the knowledge base

Knowledge base: This is a persistent knowledge repository. When answering questions, it should actively retrieve relevant background information from it. When new information related to a certain knowledge base theme is obtained (such as preference settings, decision - making processes, factual basis, or experience summaries), these contents should be gradually added to the corresponding knowledge base.

You can think of it this way: Instead of stuffing all information into one "brain" at once, Claude stores information in different "notebooks" in an organized manner.

When you ask Claude questions or work with it, it will actively search these knowledge bases to see if there is any background information related to the current task, instead of starting from scratch every time.

More importantly, these knowledge bases are not static.

When Claude learns something new about a certain topic, such as your personal preferences, decisions you've made, confirmed facts, or mistakes you've made, it will gradually add these to the corresponding knowledge base.

In short, the more you use it, the better it understands you!

The biggest change is that -

Claude no longer relies on a chaotic and uncontrollable "general memory." Instead, users can manage multiple clear and independent knowledge bases themselves.

In the future, when using Claude Cowork to do things, we can manually select which knowledge base to use as the context.

For example, use the "Project Knowledge Base" when writing a proposal, the "Workflow Knowledge Base" when running automation, and the "Data Management Knowledge Base" when organizing files.

For complex tasks such as automation and file management, this is truly revolutionary. Claude can finally truly understand "what you are doing" instead of making random guesses.

Claude Cowork takes the lead and becomes the main mode

Another clear signal is that Claude Cowork will become the main mode.

That is to say, the traditional "Chat" mode won't disappear, but it will be integrated into Cowork.

From now on, Cowork will become the default workspace for Claude in the future.

Switching of the Chat mode

You can think of it as an AI workbench that integrates chatting, files, automation, knowledge bases, and output management.

Of course, if you just want to have a casual chat, you can open the chat mode in Cowork.

Obviously, Anthropic's stance is clear: "Chatting is just the entry point. The real value lies in the workflow."

The UI is also undergoing a major overhaul. It's the era of the Artefacts on the right - hand side

Meanwhile, the UI layer of Claude is also being adjusted synchronously: There will be a dedicated Artefacts area on its right - hand sidebar, and it won't be "gone after the chat."

From now on in Claude, we can continuously generate, manage, and reuse results.

This means that Claude will place more emphasis on the sustainability of the output content in the future, rather than one - time responses.

Your relationship with Claude will no longer be just a question - answer one. It will be more like working on a project together.

More powerful automation: The MCP connector system

In addition, a keyword was mentioned in the leak: MCP Registry.

That is to say, Claude is likely to dynamically manage multiple remote connectors and install "officially approved modules" as needed.

In this way, the automation ability of Cowork will be greatly enhanced.

MCP Registry

Moreover, to complete tasks, it will automatically call the appropriate capabilities.

If this step is truly implemented, it will be very threatening, especially to tool software.

Because this means that Claude Cowork can not only help you think and write but also truly help you operate systems and tools.

Voice mode & Pixelate, the experience layer is also upgraded

In addition to the "brain upgrade" this time, Anthropic hasn't neglected the experience layer either.

One of them is the Claude Web voice mode.

According to the leak, this mode is under development. If it is officially launched, it means we can use Claude more smoothly anytime, anywhere.

Another update is the upgrade of Pixelate.

This feature allows users to convert images into pixel - art avatars. Now, it can generate higher - quality results and has been extended to the desktop application.

Overall, although these two features seem like "lightweight features," they together indicate one thing: Claude is evolving towards multi - modality and high - frequency use.

Here comes the AI colleague with permanent memory. Are you ready?

From these changes in the leak, we can clearly see Anthropic's direction -

It's not just about dialogue or simply stacking model capabilities. It's about an AI collaborator who will accompany you in your work for a long time. The keywords are knowledge base, workflow, and automation.

If previous large - language models like ChatGPT were more like on - call advisors, then Claude Cowork is more like an AI colleague who can remember things and execute tasks.

If this wave of updates is really launched, the competition among AI assistants may enter a brand - new stage from now on.

The Year of Continuous Learning

In essence, the knowledge base introduced by Anthropic this time addresses the issue of AI's continuous learning or AI memory.

Last year, Dario Amodei, the CEO of Anthropic, believed that "continuous learning will ultimately not be as difficult as it seems."

From OpenAI to Google, there is almost a consensus in the Silicon Valley tech circle: 2026 will be the year of continuous learning.

On September 5, 2024, OpenAI first opened the memory function to ChatGPT users.

This "knowledge base" of Claude Cowork seems similar to ChatGPT's memory function.