🔍 Read the full analysis: Building A Shared Second Brain 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 says he built a shared Second Brain so his AI agents can refer to and add to one long-term knowledge base across sessions. The system uses Markdown files, Git and a plain-text index; Meyer reports 1,278 pages and a 97.4% top-one recall score as of November 21, 2026.
Thorsten Meyer has described a shared memory system for his AI agents, built from Markdown files, Git and a searchable index, in a report published November 21. The system is intended to let agents retrieve and add information across separate sessions, while keeping the underlying knowledge readable and auditable by its owner.
Meyer calls the system his Second Brain. Each page holds a cluster of related facts, such as information about a project or a decision, and is stored as a plain-text Markdown file under Git version control. A single file, INDEX.md, lists each memory in one line. Meyer says the system uses no vector database or cloud service.
When an agent needs information, a script called recall.py scores the index entries first, without opening the memory pages. It then opens the highest-scoring page and returns the section that answers the question, along with the source page. Meyer says that source reference lets him check the material behind an answer. A separate script, remember.py, writes a new page, its index entry and a log entry together.
As of November 21, 2026, Meyer reports that the knowledge base contained 1,278 pages and 1,761 links, with 1,363 recalls. He gives a 97.4% score for “recall top-1,” which he defines as how often the first page selected is the right one. These figures and the performance measure are reported by Meyer; the article does not describe an independent evaluation.
The design addresses a practical limitation of session-based AI agents: information available in one session may not be present in the next. A shared store can let different agents refer to the same project details and past decisions instead of requiring the user to repeat them. In Meyer’s setup, agents can also contribute new information to that common store.
Keeping the records in ordinary files makes them easier for an owner to inspect, edit or move than memory held only inside a particular tool. Git also records changes over time. But the system’s reported recall score is not, by itself, evidence that it will work equally well for other people or tasks. Meyer describes a personal system, and says the approach requires discipline and is built for one owner, not a team.
Products Worth Considering

Ledger Nano X Bluetooth Crypto Hardware Wallet

Apple MacBook Air (2019) 13-inch, 1.6GHz Intel Core i5, 8GB RAM, 128GB SSD, Gold

Logitech MX Brio Ultra HD 4K Webcam
As an affiliate, we earn on qualifying purchases.
How the Memory System Works
Meyer frames the project as a response to agents starting each session without the user’s prior decisions, project names or knowledge of approaches that have already failed. He says he runs multiple agents and wanted one memory he could read himself, rather than separate stores that are difficult to inspect or correct.
Alongside the files and index, Meyer describes nine views for managing the system. They include a map of connections, a morning brief, a triage process for deciding whether captured information should become a memory, and a chronicle of changes. Other views show page links, live agent activity, daily operations and the services running the system. The views provide ways to monitor the collection; the core storage remains the Markdown pages and index.
The source report also describes a process for handling information left behind by agent sessions: captures can be kept as lasting memories, merged into existing pages or dropped. Meyer says the system’s index-first retrieval avoids opening every page for each query. That choice makes the quality and clarity of the one-line index entries important to how well recall works.
“One memory, shared by every agent I run, that I can open and read myself.”
— Thorsten Meyer, describing the system on ThorstenMeyerAI.com
Questions About Performance and Scale
The reported 97.4% top-one recall score is not accompanied in the source material by details about the test set, the number of questions, how correct answers were judged or the period over which the result was measured. It is therefore unclear how the figure would compare with other retrieval systems or how well it applies beyond Meyer’s own collection.
The report also does not provide a public implementation package, independent testing or details about access controls, backups and handling of sensitive information. Meyer says personal data shown in the video and images was protected, but that does not establish how the system would meet the privacy needs of other users. The article describes an individual setup; it does not show how conflicts, permissions or maintenance would be handled in a multi-user deployment.
Further Testing and Team Use
Meyer’s report presents the system as an existing personal tool and describes its operation as of November 21, 2026. It does not announce a launch date, commercial release or formal development roadmap. Readers seeking to reproduce the approach would need further practical details about the scripts and setup beyond the explanation provided in the source.
The next useful evidence would include a reproducible account of how recall accuracy is measured, examples of errors and how they are corrected, and details on how the system handles sensitive records. Whether the design can support teams, rather than one owner, also remains an open question. Until those points are addressed, the reported scale and score should be understood as Meyer’s account of his own system, not as a general performance guarantee.
Key Questions
What is Thorsten Meyer’s Second Brain?
It is a personal knowledge base shared by Meyer’s AI agents. It stores information in Markdown files, with a central index that agents use to find relevant pages.
How does an agent retrieve a memory?
According to Meyer, recall.py scores entries in the index, opens the best-matching page and returns the relevant section with its source. The described process does not search every page for each question.
How large was the system on November 21, 2026?
Meyer reports 1,278 pages, 1,761 links and 1,363 recalls as of that date. He also reports a 97.4% top-one recall score, but the source does not specify the testing method or comparison baseline.
Is the system designed for teams?
Not as described. Meyer characterizes it as a setup built for one owner and says it requires discipline. The source does not explain how it would handle team permissions or shared editing.
Does Meyer’s report prove the approach will work for other users?
No. The report describes Meyer’s system and its self-reported results. It does not provide an independent evaluation or enough test details to establish performance for other collections, users or tasks.
Source: ThorstenMeyerAI.com
