> ## Content Index > Fetch the complete content index at: https://agarhouse.com/llms.txt > Use this file to discover other available public pages before exploring further. # Meta's Muse Wonder wants to remember you. Check what you can control. - URL: https://agarhouse.com/meta-muse-wonder-memory-controls/ - Published: 2026-10-13T04:18:21.000Z - Updated: 2026-10-13T04:18:21.000Z - Description: Meta’s new AI models promise memory across conversations. Before connecting your accounts, check what you can review, delete, and limit. - Author: Gary Agar - Tags: Artificial Intelligence, News, Meta, Privacy An AI assistant that remembers your project can save you from explaining the same thing every time you open a chat. But before giving it access to more of your life, check what it remembers, what you can delete, and what it can do without asking. Meta announced Muse Wonder and Muse Wonder Full on October 12\. The company is putting memory and the ability to work through longer tasks at the center of the release. Those features could be useful. They also deserve a closer look than a benchmark chart. ## What Meta announced Meta describes Active Memory as carrying useful information across conversations and bringing it back when it matters. Its example of a capable assistant goes beyond answering a question: it should remember relevant context while helping with a larger task. The other feature is Active Focus. Meta says its models can try different approaches, work through blockers, and adapt to new information. The company also says they ask for clarification and check in before important actions. That last part matters. An assistant suggesting a change to a document and an assistant making a purchase need different permissions. Remembering what someone wants should not automatically mean having permission to act on it. ## Which version can you use? Meta says Muse Wonder is available through its Meta Model API and Muse Code. Muse Wonder Full is available to Muse Code subscribers. The company says Wonder will also power experiences in its Muse assistant, with subscribers getting access first. TechCrunch reports API pricing of $4 per million input tokens and $20 per million output tokens for Wonder. It also reports that Wonder Full is not being offered through the API. If you are building an app, check which model your integration can actually call. A feature available in a subscription product is not necessarily available through an API. ## Read the performance claims carefully Meta reports strong results across coding, memory, and other evaluations. TechCrunch also describes the models' benchmark comparisons with competing systems, based on Meta's testing. Those results are a reason to investigate, not a reason to hand over every account you own. A benchmark does not tell you whether an assistant will correctly remember an unusual requirement in your project or notice that an old preference no longer applies. This is a look at the launch announcement and reporting, not a hands-on review. The claims about how these models perform come from Meta's evaluations. ## What to check before connecting your accounts Start with the memory controls. Look for a way to review what the assistant has saved, correct information, and remove it. Check whether deleting a conversation also deletes any information stored separately as memory. Do not assume those are the same operation. Next, look at permissions. Can you let it read a calendar without changing events? Can it draft a message without sending it? Does it ask before spending money or sharing a file? Try a small task using information that is not sensitive. Give it a requirement, change that requirement later, and see whether it follows the correction. If it gets something wrong, check whether you can fix the saved information rather than repeatedly correcting the same mistake. These are checks to make before use, not claims that Muse currently provides every control on this list. Meta's launch announcement describes its approach to memory and safety, but it is not a complete guide to account permissions, retention, or deletion. A useful assistant should save you work while leaving you in charge. Start with limited access, check the results, and only expand its permissions when there is a clear reason to do so. ## Sources - [Meta: Muse Wonder launch announcement](https://research.meta.ai/blog/muse-wonder?ref=agarhouse.com) - [TechCrunch: Meta releases Muse Wonder](https://techcrunch.com/2026/10/12/meta-releases-its-first-frontier-model-muse-wonder/?ref=agarhouse.com)