🔍 Read the full analysis: What A Second Brain Means For AI Agents on ThorstenMeyerAI.com
Get hardware and tech essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Thorsten Meyer describes a personal knowledge base that lets multiple AI agents read and add to the same long-term memory. Built from Markdown files, a Git history and an index, it reportedly held 1,278 pages and achieved 97.4% top-1 recall as of Nov. 21, 2026; those performance figures are the author’s own measurements.
Thorsten Meyer has described a shared long-term memory system for the AI agents he uses, allowing them to consult and add to one personal knowledge base rather than start each session without prior context. In an article published Nov. 21, 2026, Meyer said the system contained 1,278 pages and recorded a 97.4% top-1 recall score—a measure he defines as how often its first selected page is the right one.
Meyer’s system, which he calls a Second Brain, stores information in plain Markdown files maintained in Git. Each page groups related facts, such as details about a project or a decision. A single index file, INDEX.md, lists the pages so an agent can search the catalogue before opening a relevant file. Meyer says the setup uses no vector database or cloud service.
When an agent asks for information, a script called recall.py scores the index entries, opens the highest-ranked page and returns the section most relevant to the question. Meyer says the response includes its source page, making it possible for him to check the underlying note. A separate script, remember.py, writes a new page, its index entry and a log entry together.
That combined write is meant to keep the catalogue aligned with the stored notes. Meyer reported 1,761 links among the pages and 1,363 recalls as of Nov. 21. These figures and the 97.4% score come from Meyer’s account; the article does not provide an independent evaluation or detailed testing method.
The system addresses a practical limitation of session-based AI use: information an agent can see in one conversation may not carry into the next. A shared store can let different agents reuse project details, prior decisions and lessons from failed approaches, reducing the need for a person to repeat that context.
Its plain-text files and source-linked answers also offer a way to inspect and correct what agents use as memory. That matters when an answer depends on a stored fact: a user can open the cited page rather than rely only on an opaque recollection. Git history can show changes over time, according to Meyer.
The approach has trade-offs. It depends on keeping pages and index entries useful, and Meyer says the current arrangement is designed for one owner, not a team. The article does not establish that this design will perform as well for other users, larger shared workspaces or different kinds of agent tasks.
Products Worth Considering

Bose Lifestyle Ultra Wireless Home Speaker, White Smoke

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite

Sony WH-1000XM6 Wireless Noise-Cancelling Headphones – Midnight Blue
As an affiliate, we earn on qualifying purchases.
From Session Context to Stored Notes
An AI agent typically works from information available in its active context window. When a session ends, that working context may no longer be available to the next session. Meyer says that led him to build a single memory that all his agents could consult and update, instead of relying on separate memory features tied to individual tools.
The article describes more than the underlying files. Meyer lists nine views for monitoring the system, including a Map of connections, a daily Brief, a Triage process for deciding which session captures become lasting notes, and a Chronicle of changes. Other views track activity, page links and the services running the system. These are presented as interfaces around the file-based memory, not as separate proof of its reported recall performance.
“I wanted the opposite. One memory, shared by every agent I run, that I can open and read myself.”
— Thorsten Meyer, writing on ThorstenMeyerAI.com
Limits of the Reported Recall Score
The 97.4% top-1 score is a figure reported by Meyer, and the source does not describe the test set, number of questions, scoring procedure or independent validation. It is not clear how the system performs on unfamiliar queries, outdated information, conflicting notes or a substantially larger knowledge base.
The article also does not give a full account of access controls, backup arrangements or how sensitive information would be protected if the system were adapted for several users. Meyer says the example is built for one owner and that personal information shown in the accompanying video and images is protected. The source does not establish what safeguards another deployment would need.
How the System May Be Evaluated
Meyer’s article describes the system’s current design and measurements as of Nov. 21, 2026, but does not announce a release schedule, public benchmark or team version. Further evidence about its usefulness would depend on clearer testing details, including how recall is measured and how it behaves when notes overlap or change.
For now, the described system remains Meyer’s account of a personal workflow. Readers considering a similar setup would need to decide how to organise and review stored information, and how to manage access and sensitive data. The source does not say whether Meyer plans to publish the scripts or extend the system beyond his own agents.
Key Questions
What does Meyer mean by a Second Brain for AI agents?
It is a shared knowledge base that Meyer says his agents can search and update across sessions. It stores notes as Markdown files and uses an index to help find relevant pages.
How does an agent retrieve information?
According to Meyer, recall.py scores lines in the index, opens the best-matching page and returns the relevant section with a source reference. The account does not provide independent testing of that process.
What does the 97.4% figure measure?
Meyer defines top-1 recall as how often the first page selected by the system is the right one. The article does not specify the test sample or method, so the figure should be read as his reported measurement.
Can this system be used by a team?
Meyer says his setup is built for one owner, not a team. The article does not describe shared-user permissions or a team deployment.
Does the system use a cloud service or vector database?
Meyer says it uses plain Markdown files kept in Git, with a single index, and no vector database or cloud service. Those details describe his system, not every possible second-brain design.
Source: ThorstenMeyerAI.com
