🔍 Read the full analysis: From Separate Agents To Shared Memory: My Second Brain on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Thorsten Meyer says he built a shared “Second Brain” so the AI agents he uses can consult and update one long-term knowledge base instead of starting each session without context. The system uses Markdown files, Git and a custom index, rather than a vector database or cloud service; Meyer reports 1,278 pages and a 97.4% top-one recall score as of Nov. 21, 2026. Those figures and performance claims come from Meyer, and the article does not provide an independent evaluation.
Thorsten Meyer says he has built a shared long-term memory for the AI agents he uses, allowing them to look up and add information across sessions instead of repeatedly starting without knowledge of his work. In an article published November 21, 2026, Meyer described the system, which stores information in Markdown files tracked with Git and reports 1,278 pages and a 97.4% top-one recall score.
Meyer calls the setup his Second Brain. Its knowledge base consists of plain-text Markdown pages, each covering a cluster of facts, alongside an index file named INDEX.md. He says there is no vector database or cloud service in the design. Git keeps a history of changes, while the files can be opened and edited with ordinary tools.
When an agent needs information, a script called recall.py scores the index entries first, then opens the best-matching page and returns the relevant section with its source. Meyer says this lets him check the page behind an answer. A separate script, remember.py, writes a new page, its index entry and a log line together, a workflow he says helps keep the catalogue aligned with the stored notes.
Meyer reports that the system had 1,278 pages, 1,761 links and 1,363 recalls as of Nov. 21. He gives its top-one recall score—the system’s own measure of how often its first selected page is the right one—as 97.4%. The article does not describe an independent test, the evaluation set or the period over which the score was calculated.
A More Inspectable Agent Memory
The system addresses a practical limitation of session-based AI agents: information available in one session may not carry into the next. A shared store could reduce repeated explanations and let different agents draw on the same project details, decisions and working preferences. For people who use multiple agents, that is the main proposition of Meyer’s design.
Its emphasis on readable files and source-linked recall also offers a different trade-off from memory features tied to a particular application. A person can inspect or edit the underlying notes and, in principle, move the files using standard tools. Meyer’s reported 97.4% score is a performance claim about his own system, not evidence that the approach will work equally well for other users or tasks.
The approach also places responsibility on the owner. Meyer describes it as a personal system, not a team setup; shared access, permissions, sensitive information and coordination among multiple contributors are not addressed in the account. Its usefulness depends on maintaining accurate notes and an index that agents can search effectively.
Products Worth Considering

Samsung Galaxy Tab A11+ Kids Edition, 128GB, Royal Blue

Jackery HomePower 3000 Portable Power Station, 3072Wh, 3600W

Kindle Paperwhite Signature Edition 32GB (2024 Model) – Metallic Black
As an affiliate, we earn on qualifying purchases.
AI agents generally respond using information available to them during a session, including the prompt and any files or tools they can access. Meyer’s stated problem is that when a session ends, an agent may not retain the details needed in a later conversation. He says he uses many agents and wanted one store that each could consult and update.
The system described in the article is organized around an index and linked text pages, rather than a separate retrieval database. Meyer also describes nine views for monitoring or managing it, including a daily summary, a triage process for deciding which session notes become lasting memories, a change log, a graph of linked pages and a console for services and machines. These are features of his implementation, not independent evidence of its accuracy or reliability.
The article presents a personal build and its reported measurements, rather than a product launch or a controlled comparison with other memory systems. Meyer says the design prioritizes readability and traceability, while acknowledging the discipline required to keep it useful.
“I got tired of being their memory. So I built them one.”
— Thorsten Meyer, describing his motivation
How the Recall Score Was Measured
The article does not explain how Meyer calculated the 97.4% top-one score, what counted as a correct page, how many queries were tested or whether the evaluation was independent. It also does not provide a comparison with other retrieval methods, so the figure cannot establish how the system performs outside its own measurement process.
Other practical details remain unspecified, including how sensitive information is protected, how permissions would work if more than one person used the system, and how the files and index are backed up or synchronized across machines. Meyer’s description says the system uses local-style files and Git, but does not fully detail its operating environment. The article also does not establish how much time upkeep requires or how well the approach works for different types of work.
Testing the Design Beyond One Owner
Meyer’s article describes the system as it stood on November 21, 2026, and offers an account of its structure and reported activity. It does not announce a public release, a team version or a scheduled next milestone. The next useful evidence for readers would be a fuller explanation of the recall test, along with examples of where the system succeeds or fails.
For anyone considering a similar setup, the immediate next step would be to judge whether a searchable set of editable files fits their work and whether they can keep it current. Meyer’s account makes inspectability and shared access the design goals; it does not establish that this specific setup is ready to adopt broadly. Whether the idea scales beyond one owner remains an open question.
Key Questions
What is Thorsten Meyer’s Second Brain?
It is a personal knowledge base Meyer says he shares with the AI agents he uses. Agents can look up information in it and add new entries.
How does the system store information?
Meyer says it uses Markdown files tracked with Git, plus an index file listing the memories. He says it does not use a vector database or cloud service.
What does the 97.4% figure mean?
Meyer describes it as the system’s top-one recall score: how often the first page selected is the right one. The article does not give the test method or an independent evaluation.
Can the system be used by a team?
Meyer characterizes it as a setup built for one owner, not a team. The article does not describe team permissions or collaboration features.
How large was the knowledge base?
As of November 21, 2026, Meyer reported 1,278 pages, 1,761 links and 1,363 recalls.
Source: ThorstenMeyerAI.com
