🔍 Read the full analysis: A Second Brain Lets My AI Agents Share What They Know on ThorstenMeyerAI.com
Get the latest gadgets 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 says he built a personal knowledge base that lets his AI agents share information across sessions. It uses markdown files, git and a script that searches an index; Meyer reports 1,278 pages and a 97.4% top-1 recall score as of Nov. 21, 2026. The system is designed for one owner, and its performance has not been independently verified in the source material.
Thorsten Meyer has described a shared memory system for the AI agents he runs, built from markdown files and a searchable index rather than a vector database or cloud service. In an article published Nov. 21, 2026, Meyer says the system lets agents retrieve and add information across sessions, addressing the problem of repeatedly briefing agents that otherwise begin each session without knowledge of prior work.
Meyer calls the project his Second Brain. Each memory is stored in a plain-text page, with each page covering a cluster of related information, such as a project or decision. The files are kept in git, which Meyer says records changes over time. A separate file, INDEX.md, lists each memory in a single line and serves as the starting point for searches.
When an agent needs information, a script called recall.py scores the index lines, opens the page it considers the best match and returns the relevant section with its source. Meyer says this allows a user to check where an answer came from. A second script, remember.py, writes a new page, its index entry and a log entry together, a process intended to keep the index aligned with the stored pages.
Meyer reports that, as of Nov. 21, 2026, the system contained 1,278 pages and 1,761 links, and had recorded 1,363 recalls. He gives a 97.4% top-1 score, which he defines as how often the first page selected by the system is the right one. These are figures reported by Meyer; the article provides no independent evaluation or detailed test methodology.
The system’s main practical aim is to reduce repeated setup: an agent can consult information from previous work rather than relying only on what is present in its current session. For people using several agents or switching tools, a common memory store could make project decisions and terminology easier to carry between sessions. It also gives the owner a place to correct a fact once rather than updating separate tool-specific memories.
Meyer’s design puts readability and traceability ahead of automated retrieval infrastructure. Users can inspect the files directly, review changes in git and follow citations back to a source page. That may help users verify recalled information, though it does not by itself establish that memories are accurate or that the system will choose the right page in every case.
The trade-off is scope and upkeep. Meyer describes the arrangement as built for one owner, not a team, and says it depends on discipline. People considering a similar setup would need to decide who can change records, how sensitive information is handled, and how to resolve conflicting or outdated entries.
Products Worth Considering

Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite

Sonos Era 100 SL Compact Microphone-Free Speaker, Black

Jackery HomePower 3600 Plus Portable Power Station
As an affiliate, we earn on qualifying purchases.
From Session Notes to One Index
Meyer frames the project around a limitation of session-based AI use: an agent can use the information available in its current context, but that context does not automatically carry over when a session ends. Some tools offer memory features, he writes, but those memories may be difficult to inspect or edit and can be separate across products.
His alternative is a folder of text files with one index that all of his agents can consult. The article also describes nine views for monitoring and reviewing the system, including a map of connections, a daily brief, a triage process for session captures, a change log, an explorer of linked pages and consoles for live activity and services. These are features Meyer says he built around the underlying files; the source does not provide an outside assessment of their operation.
“I got tired of being their memory. So I built them one.”
— Thorsten Meyer, writing on ThorstenMeyerAI.com
Performance and Team Use Remain Open
The source does not explain how Meyer calculated the 97.4% top-1 score, including the size and composition of the test set, how a correct match was defined, or whether the score was measured over a particular period. The reported page, link and recall totals are also self-reported and are not accompanied by an independent audit.
It is not clear how the system handles conflicting, stale or incorrect memories, or how it controls access if several people or agents write to the same store. Meyer says the setup is for one owner, so its suitability for team use has not been established. The article also does not provide a full account of security safeguards, backup procedures or the risks of placing sensitive information in the files.
How the System May Develop
Meyer’s article presents the Second Brain as an existing personal system and includes an explainer video showing its views and workflows. It does not announce a product release, public service, team version or planned launch date. The next developments are therefore unclear from the material provided.
Readers evaluating the approach can look for further detail on the recall test, data safeguards and the process for reviewing memories before they become part of the long-term record. Any future performance figures would be easier to compare if Meyer published the test method and its time period alongside the result.
Key Questions
What is Thorsten Meyer’s Second Brain?
It is a personal knowledge base that Meyer says he shares with the AI agents he runs, so they can retrieve and add information across sessions.
How does an agent find a memory?
According to Meyer, recall.py scores one-line descriptions in INDEX.md, opens the best-matching page and returns the relevant section with its source.
How large was the system in November 2026?
Meyer reported 1,278 pages, 1,761 links and 1,363 recalls as of Nov. 21, 2026. Those figures come from his article and have not been independently verified in the source material.
Is the reported 97.4% score independently verified?
The source gives no independent evaluation or full testing method. Meyer describes the figure as a top-1 recall score, meaning how often the first page selected is the right one by his system’s measure.
Is the system built for teams?
Meyer says the setup is designed for one owner, not a team. The source does not describe shared access controls or a team deployment.
Source: ThorstenMeyerAI.com
