Why Shared Memory Matters For My AI Agents
KIDieser Beitrag wurde mit Unterstützung künstlicher Intelligenz (KI) erstellt.

🔍 Read the full analysis: Why Shared Memory Matters For My AI Agents on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

TL;DR

Thorsten Meyer says he built a shared “Second Brain” so his AI agents can consult and update one long-term knowledge base rather than start each session without prior context. The system uses Markdown files, a Git-tracked index and scripts; Meyer reports 1,278 pages and 97.4% top-1 recall accuracy as of Nov. 21, 2026. The figures are his own, and the setup is designed for one owner rather than a team.

Thorsten Meyer says he has built a shared long-term memory system for the AI agents he uses, allowing them to consult and add to the same knowledge base across sessions. In an article published Nov. 21, 2026, he described a setup made of Markdown files, a Git-tracked index and small scripts, and reported that it contained 1,278 pages with a 97.4% top-1 recall score on that date.

Meyer calls the system his “Second Brain.” Each page groups related information, such as details about a project or a decision. An index file, INDEX.md, lists each memory in a single line. When an agent needs information, a script called recall.py scores those index entries, opens the best-matching page and returns the relevant section with its source. Meyer says the design does not use a vector database or cloud service.

New information enters through remember.py, which writes a page, its index entry and a log entry together. Meyer says this process is intended to prevent the index from becoming separated from the notes it describes. Because the files are stored in Git, he says changes can be traced through the repository’s history and the system’s log.

As of Nov. 21, 2026, Meyer reported 1,278 pages, 1,761 links and 1,363 recalls. He said the system’s “top-1” measure—how often its first selected page is the right one—was 97.4%. These are figures supplied by Meyer; the article does not describe an independent test or provide the underlying evaluation data.

At a glance
reportWhen: Published November 21, 2026; system fig…
The developmentThorsten Meyer published a description of a shared memory system for his AI agents, detailing its file-based design and reporting its size and recall performance.

A Shared Record Across Agent Sessions

The system addresses a practical limitation Meyer identifies: an agent’s session context does not automatically carry into the next session. Without a shared record, users may have to repeat project details, previous decisions and failed approaches. A common memory could let different agents use the same information, rather than keeping separate records inside individual tools.

Meyer also emphasizes readability and traceability. Since the memories are plain files, he can inspect or edit them directly, and an agent’s answer can point back to a source page. That could make mistakes easier to find than in a memory store that users cannot readily view. This is Meyer’s account of the benefits; the article does not compare the system with other products or report results from independent users.

The approach has trade-offs. It relies on a well-maintained index and on deciding which session information merits lasting storage. Meyer says the system is built for one owner, not a team, and identifies discipline as a cost. Shared access alone does not settle questions such as permissions, conflicting edits or how personal data should be managed in a group.

Products Worth Considering

Apple Mac mini with M2 chip, 16GB unified memory, 256GB SSD, Silver

Apple Mac mini with M2 chip, 16GB unified memory, 256GB SSD, Silver

Canon IVY 2 Mini Photo Printer Bundle with 110 ZINK Sticky-Back Prints

Canon IVY 2 Mini Photo Printer Bundle with 110 ZINK Sticky-Back Prints

As an affiliate, we earn on qualifying purchases.

From Session Context to Stored Notes

Meyer frames the project around the gap between an AI agent’s current context and information that must persist over time. He says some tools offer memory features, but those can be difficult to inspect, edit or audit and may differ from one tool to another. His stated aim was a single memory shared across the agents he runs, with answers linked to material he can check.

The system has nine named views for monitoring and managing that information. Meyer describes tools for mapping connections, reviewing new captures, checking frequently used pages, viewing the change history and watching agents recall or write memories. A daily brief summarizes the previous day, while a triage process sorts session captures into items to keep, merge or discard. These views sit on top of the files and index; Meyer says the underlying storage remains a folder of text files.

His article and accompanying explainer video present this as a personal working system, not a general-purpose product announcement. Meyer says personal information shown in the video and images has been protected or masked. The source does not offer a detailed account of the system’s software dependencies, security arrangements or the process used to validate its reported metrics.

““One memory, shared by every agent I run, that I can open and read myself.””

— Thorsten Meyer, writing on ThorstenMeyerAI.com

How the Recall Score Was Tested

The reported 97.4% top-1 score is not accompanied in the source material by a test set, sample size, evaluation period or definition of a correct match beyond Meyer’s description. It is therefore unclear how the score was calculated or how well it would represent performance on other users’ information and questions. The 1,363 recalls are reported as a count, not as a measure of accuracy or a rate over a specified time window.

The article also does not explain how the system handles sensitive data, access controls, simultaneous changes or recovery if files are lost. Meyer describes a setup for personal use, so its performance and maintenance requirements in a shared workplace remain unestablished. No independent audit or comparison with other memory systems is included.

More Detail on Use and Evaluation

Meyer’s article points readers to an explainer video that walks through the system’s nine views and shows recall and writing activity. Further reporting would be needed to verify the reported performance, understand the evaluation method and establish how the approach works beyond a single-owner setup.

For readers considering a similar design, the immediate questions raised by Meyer’s account are how to maintain the index, review session captures and protect stored information. The source does not announce a product release, rollout schedule or planned independent test, so no such next step is confirmed.

Key Questions

What is Meyer’s Second Brain?

It is Meyer’s shared knowledge base for the AI agents he uses. Agents can look up information in it and add memories, according to his article.

How does an agent find a memory?

Meyer says recall.py scores lines in an index first, then opens the best-matching page and returns the relevant section with its source.

How large was the system on Nov. 21, 2026?

Meyer reported 1,278 pages, 1,761 links and 1,363 recalls as of that date. Those are source-reported counts, not independently verified figures.

Is the 97.4% recall score independently verified?

The source provides no independent verification or detailed testing method. Meyer describes it as the system’s top-1 score, or how often its first selected page is the right one.

Can teams use the same setup?

Meyer says his setup is built for one owner, not a team. The source does not explain team permissions or how it would handle shared editing.

Source: ThorstenMeyerAI.com

This content is for general information only and is not financial, tax or legal advice. Consult a qualified professional for decisions about your money.
You May Also Like

Choosing A Content Creator Laptop: 6 Picks For AI, Editing, And Design In 2027

A source-based comparison of six creator laptops, their stated specs, tradeoffs and the claims that still need independent testing.

7 AI Trends That Will Define 2026

Explore the seven key AI trends expected to define 2026, including advances in generative AI, ethical frameworks, and industry-specific applications, based on expert analysis.

Claude Code At Anthropic: Tracking Three Team Members’ Departures

Analytics India Magazine reported three Claude Code-associated departures at Anthropic within 10 days, but did not identify the people or their roles.

Technology Operations Signal Monitor: How Google Helped Destroy Adoption Of RSS Feeds (2023)

A recent analysis shows how Google’s platform and tooling changes contributed to the decline of RSS feed adoption, impacting small software companies.