🔍 Read the full analysis: How I Connect AI Agents Through Shared Memory 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 shared “Second Brain” so his AI agents can access and update the same long-term memory across sessions. As of November 21, 2026, it contained 1,278 pages, and Meyer reports a 97.4% top-one recall score; the figures and performance measure come from his own account.
Thorsten Meyer has described a shared memory system that lets the AI agents he runs consult and update one personal knowledge base, rather than begin each session without details from previous work. In a post published November 21, 2026, Meyer said the system, which he calls his Second Brain, held 1,278 pages and had recorded 1,363 recalls as of that date. The project offers a concrete example of how one operator is addressing continuity across agents, though its reported performance has not been independently verified.
The system stores information in Markdown files tracked in Git, with one page for each cluster of related facts, such as a project or decision. A single file, INDEX.md, lists each memory in one line. Meyer says the setup has no vector database or cloud service and can be opened in an ordinary text editor. Git records changes over time, giving the owner a history of edits.
When an agent needs information, a script called recall.py scores the index lines first, then opens the page it considers the best match. It returns the relevant section along with its source, Meyer says. For adding a memory, remember.py writes the page, its index entry and a log entry together. Meyer says this coordinated step is intended to prevent the index from falling out of sync with the pages.
Meyer reports that the collection contained 1,278 pages and 1,761 links on November 21, 2026. He also says it had completed 1,363 recalls and scored 97.4% on “top-1” recall, his measure of how often the first page selected was the right one. Those are figures from the author’s description; the post does not provide an independent evaluation or enough detail to compare the score with other memory systems.
The system addresses a practical limitation of session-based AI agents: information in an active context may not carry into the next session. A shared store can give different agents access to the same project details, decisions and preferences, potentially reducing repeated explanations and inconsistent answers.
Meyer’s design also emphasizes inspectability and control. Because the memories are plain files and answers point to a source page, a person can review or edit the material rather than rely entirely on a tool-specific memory feature. This may appeal to people who want to keep records portable and auditable, but it shifts responsibility to the owner to maintain accurate, useful notes.
The author describes a personal setup, not a validated team product. Its reported figures show the scale of his own use, but do not establish that the same approach will perform equally well for other users, workloads or collections.
Products Worth Considering

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

Amazon Kindle Scribe (16GB) with Premium Pen, Tungsten

Jackery Fridge Guard 1024Wh Backup Power Station with Mount
As an affiliate, we earn on qualifying purchases.
From Session Context to Stored Notes
Meyer frames the project around the gap between an agent’s current context window and information needed in later sessions. Some software offers built-in memory, he writes, but he wanted a shared store that he could read and correct, rather than separate memory tied to each tool.
The workflow includes more than lookup and storage. Meyer lists nine views for monitoring and organizing the system, including a map of connections, a daily brief, a triage process for session captures, a change log, a graph explorer and a live view of agent activity. These are presented as parts of his implementation; the core memory itself is the folder of pages and index.
In the post, Meyer says a triage step determines whether information captured during sessions should become a lasting memory, be merged into an existing page or be discarded. That process matters because a shared store is only useful if it remains current and avoids accumulating conflicting or low-value entries.
“One memory, shared by every agent I run, that I can open and read myself.”
— Thorsten Meyer, describing the system on ThorstenMeyerAI.com
Limits of the Reported Results
The reported 97.4% top-one score is not accompanied in the supplied article text by a test set, scoring procedure, error rate or independent assessment. It is therefore unclear how the score was calculated or how well it would transfer to other systems and collections. The post also does not give enough information to assess how often stored facts become outdated or how conflicts are handled beyond the described triage process.
Meyer says the setup is designed for one owner, not a team. The source does not detail access controls, backup arrangements, privacy safeguards or how concurrent edits by multiple agents are managed. It also does not quantify the time required to maintain the system or compare its performance with commercial memory tools.
How the Memory Is Maintained
Meyer describes recurring maintenance through a daily brief, triage of session captures and an overnight review, alongside tools that monitor the system’s services. The immediate next step for the project, as presented in the post, is continued operation and upkeep of its pages, index and connected tools.
Readers considering a similar approach can evaluate the design against their own needs: whether they want plain, portable files; who will check and update memories; and how they will handle access, backups and sensitive information. Meyer’s account provides a detailed implementation example, but broader claims about accuracy or suitability would require testing beyond the figures he reports.
Key Questions
What is Meyer’s Second Brain?
It is a personal knowledge base that Meyer says his AI agents can share. It stores memories in Markdown files and uses an index to help agents find relevant pages.
How do the agents find a memory?
According to Meyer, recall.py scores entries 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 and 1,761 links as of November 21, 2026. He also reported 1,363 recalls; these are figures from his own account.
Is the 97.4% recall score independently verified?
The supplied post does not describe an independent evaluation or provide enough methodology to reproduce the score. It should be treated as Meyer’s reported measure.
Is the setup intended for teams?
No. Meyer characterizes the system as designed for one owner, not a team. The post does not provide details about team permissions or multi-user editing.
Source: ThorstenMeyerAI.com
