AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: How AI Agents Can Work From The Same Memory on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get smart everyday buys delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

Thorsten Meyer describes a personal knowledge base that lets multiple AI agents read and add to the same long-term memory. His setup uses Markdown files and Git, with a script that searches an index before opening a relevant page. Meyer reports 1,278 pages and a 97.4% top-one recall score as of Nov. 21, 2026; those figures describe his own system, not an independently tested benchmark.

Thorsten Meyer has described a shared long-term memory that lets the AI agents he runs look up and add information across separate sessions. In a report published Nov. 21 on ThorstenMeyerAI.com, he says the system uses Markdown files stored in Git and a searchable index, rather than a vector database or cloud service.

Meyer calls the knowledge base his “Second Brain.” Each page holds a cluster of related facts, such as information 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.

That source reference is a central feature of the design, Meyer says: users can inspect the page behind an answer and correct information in the underlying file. Git records changes over time. A separate script, remember.py, writes a new page, its index entry and a log entry together, a process Meyer says is intended to keep the index aligned with the stored notes.

Meyer reports that the system held 1,278 pages and 1,761 links on Nov. 21, 2026, and had recorded 1,363 recalls. He also reports a 97.4% top-one recall score, which he defines as how often the first page selected is the right one. These are figures he provides for his own system; the article does not describe an independent evaluation or a comparison with other memory tools.

At a glance
reportWhen: Published November 21, 2026
The developmentThorsten Meyer published a description of a shared, file-based memory system that his AI agents use across sessions.

A Shared Record Across Agent Sessions

A shared store can reduce the need for users to repeat project details, prior decisions and failed approaches every time an agent session starts. It also offers a way for different agents to consult the same record, rather than relying on separate memory features tied to individual tools.

The design emphasizes inspectability: information is kept in files a person can read and edit, and recalled answers can point back to their source. That may make errors easier to trace than in systems whose stored memory is difficult to examine. But the reported accuracy figure is Meyer’s own measure, and the article does not establish how the approach performs on other users’ data or tasks.

The trade-off is maintenance. Meyer says the setup is built for one owner, not a team. Shared files and an index do not by themselves settle questions such as who may change a memory, how conflicting updates are handled, or how sensitive information should be controlled when agents can write to the same store.

Products Worth Considering

Vari ComfortEdge 48x30 Inch Adjustable Electric Standing Desk

Vari ComfortEdge 48×30 Inch Adjustable Electric Standing Desk

As an affiliate, we earn on qualifying purchases.

From Session Notes to Long-Term Memory

The system addresses a limitation Meyer identifies in agent sessions: an agent works with information available in its current context window, and that context does not automatically carry over when the session ends. Some tools provide memory features, he writes, but they may be hard to read or audit and can differ from one tool to another.

Meyer’s alternative is a folder of text files and an index, with scripts governing how agents retrieve and add information. He also describes nine views for monitoring the system, including a map of links, a daily brief, a triage process for session captures, a change log and live monitoring of recalls and writes. Those interfaces sit on top of the files; the underlying store remains plain text tracked in Git, according to his account.

The article presents this as a working personal system, not a general product launch. Its figures and descriptions reflect Meyer’s own implementation and are not independently verified in the source material.

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

— Thorsten Meyer, describing the system on ThorstenMeyerAI.com

Limits of the Reported Accuracy

The source does not provide an independent test of the 97.4% top-one score, the questions used to calculate it, or the period over which the 1,363 recalls occurred. It is therefore unclear how the score would compare with other retrieval systems or hold up on a different memory collection.

The article also does not detail access controls, privacy protections, backup and recovery procedures, or how simultaneous changes from multiple agents are resolved. Meyer says the setup is designed for one owner, leaving its suitability for a team environment unestablished. The source material’s final description of the system’s figures is cut off, so no additional metrics can be confirmed from it.

How the System May Develop

Meyer’s article points readers to an explainer video showing the system’s views and workflows, but it does not announce a public release, product roadmap or formal evaluation. The next useful evidence would be details on the recall test method, results across varied tasks, and how the memory handles incorrect or conflicting entries.

For now, the described setup remains a personal implementation. Any wider use would need clearer information about permissions, team editing and protection of stored data. Meyer’s account shows how a shared memory can be assembled from ordinary files and scripts, while leaving its broader reliability and scalability open.

Key Questions

What is Meyer’s Second Brain?

It is a personal knowledge base that Meyer says his AI agents can consult and add to across sessions. It stores related facts in Markdown pages and lists those pages in an index.

How do the agents find information?

According to Meyer, recall.py scores lines in the index first, then opens the best-matching page and returns the relevant section with a source reference. The described process does not search every page directly.

Does the system use a vector database or cloud service?

Meyer says it uses plain Markdown files stored in Git and does not use a vector database or cloud service.

What does the 97.4% figure measure?

Meyer defines it as the rate at which the first page selected by the system is the right one. The source does not give an independent test, detailed methodology or comparison baseline.

Can a team use the same setup?

The article does not establish that. Meyer describes the system as built for one owner, not a team, and does not explain team permissions or how concurrent edits are managed.

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

AI output review queue for customer support macros

Support teams are testing a new AI output review queue for customer support macros to ensure policy compliance and tone accuracy before publication.

FAA Grounds Flights Nationwide For Hours: What Happened?

A nationwide FAA ground stop lasted about four hours on Nov. 13 after communications problems. Officials have not published a written root-cause report.

Business Owners Can Practice Speech Skills Before Facing The Camera

IdeaNavigator AI outlines a practice app for business owners who freeze on camera, with a 30-day cohort proposed to test whether it helps.

Thinking Of Leaving Claude? Consider These Switching Costs

A report says Meta and Microsoft reduced internal Claude use. Their alternatives show why switching models can cost enterprises more than tokens.