TL;DR
Thorsten Meyer says he built a shared knowledge base so his AI agents can retrieve and add information across sessions. The system uses Markdown files, a single index and scripts; he reports a 97.4% top-one recall score, but the figure comes from his own system and its evaluation method is not detailed in the source.
Thorsten Meyer published a description of a shared memory system for the AI agents he runs, using Markdown files, a single index and scripts to carry information between sessions. He reports that the system had 1,278 pages and a 97.4% top-one recall score as of November 21, 2026; those figures are his own measurements, and the article does not detail how the score was independently assessed.
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Meyer’s system, which he calls a Second Brain, is intended to address a recurring limitation: an agent’s active context ends with its session, leaving it without prior decisions or project details unless those are supplied again. The knowledge base is shared across the agents he uses, so they can look up information stored in one place rather than relying on separate memories for each tool.
Each memory is kept in a plain Markdown page in Git, with pages organized around clusters of facts such as a project or decision. An index file, INDEX.md, lists each memory in a single line. For a question, Meyer says, the recall.py script scores those index lines, opens the best-matching page and returns the relevant section with its source. The design does not use a vector database or cloud service, according to his account.
New information is added through remember.py, which writes the memory page, its index entry and a log line together. Meyer says this keeps the index aligned with the notes when the system is used through that normal process. He reports 1,761 links and 1,363 recalls as well as the page count. The explainer also describes nine views of the system, including a map of connections among projects, skills, routines and the apps that interact with the memory.
A Second Brain for AI Agents in Sync
Thorsten Meyer describes a shared, file-based memory that lets his AI agents retrieve and add information across sessions.
One shared record
The same inspectable knowledge base is available to the AI agents Meyer runs.
Markdown pages
Notes in Git group facts around projects, decisions and other topics.
One list to search
INDEX.md lists each memory page in a single line.
Find and add
recall.py finds a relevant page; remember.py writes the page, index entry and log line together.
How a recall works
recall.py ranks the lines in INDEX.md against a question.
The script opens the matching Markdown page.
It returns a relevant section with its source.
What the account establishes
- Less repeated setup across sessions
- Readers can inspect and correct stored facts
- Earlier decisions can remain available
- The system is described for one owner
- Team access and simultaneous edits are unresolved
- Handling of conflicting or outdated facts is not detailed
Read the score carefully
A shared, inspectable memory could reduce repeated setup when a person works with several AI agents. Meyer’s approach also makes it possible to check the source behind a recalled answer and edit a fact in the underlying files. That is relevant for work where earlier choices, project terminology or failed approaches need to remain available beyond one chat session.
The trade-off in Meyer’s description is that a system based on plain files depends on careful organization and upkeep. He frames it as a setup for one owner rather than a team. The reported 97.4% top-one score suggests the first result was often the correct page by his measure, but without a published evaluation procedure or outside validation, readers cannot compare it directly with other memory systems.
Products Worth Considering

Amazon Kindle Scribe Colorsoft 64GB with Premium Pen

LG 27GS60QC-B Ultragear 27-inch Curved Gaming Monitor (QHD, 180Hz, 1ms, 1000R)

Apple AirPods Pro (2nd Generation) Wireless Earbuds with MagSafe Charging Case (Lightning)
As an affiliate, we earn on qualifying purchases.
From Chat Context to File Memory
AI agents typically answer using information available in their active context. Meyer says that when a session ends, that context is gone, while built-in memory features may be difficult to inspect or may differ between tools. His stated aim was to keep one memory he could read and correct, and share it across the agents he runs.
The article is a first-person account published by Thorsten Meyer AI on November 21, 2026, alongside an explainer video. The figures and design details are attributed to Meyer and his own system. The source material describes the workflow and its reported scale, but does not provide independent testing or a comparison against other approaches.
““Any agent can look something up in it, and any agent can add to it.””
— Thorsten Meyer, describing how agents use the system
How Recall Performance Was Measured
The source does not explain the test set, scoring procedure or number of evaluated questions behind the 97.4% top-one figure. It is therefore unclear how the result would hold for different kinds of questions, larger collections or other users. The account also does not describe how the system handles conflicting or outdated facts, access controls, or simultaneous edits. Meyer says the setup is designed for one owner, leaving its suitability for team use unresolved.
Further Details on Use and Scale
Meyer’s explainer video walks through the system’s nine views, according to the article. The published account gives a snapshot of its size and reported recall performance as of November 21, 2026; it does not state a later release, independent evaluation or team deployment. Any further assessment will depend on more detail about the scoring method and how the system performs as its memory grows.
Key Questions
What is Meyer’s Second Brain?
It is a personal knowledge base that Meyer says he shares with the AI agents he runs, so information can carry across sessions.
How does an agent find a memory?
According to Meyer, recall.py scores the lines in INDEX.md, opens the best-matching Markdown page and returns the relevant section with its source.
What does the 97.4% score mean?
Meyer calls it the system’s top-one recall: how often the first page selected is the right one, by the system’s own measure. The article does not explain the evaluation method.
Can a team use the system?
The account describes a system built for one owner. It does not establish how it would handle team access, permissions or simultaneous changes.
Source: Thorsten Meyer AI
