What Is A Second Brain? How AI Agents Share Memory
AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

Thorsten Meyer published an account of a personal knowledge base that lets his AI agents share long-term memory stored in markdown files and git. He reports 1,278 pages and 97.4% top-1 recall as of Nov. 21, 2026; those figures come from his own system, and the article does not provide an independent evaluation.

Thorsten Meyer published a description of a shared memory system for his AI agents on Nov. 21, 2026, outlining how plain markdown files, a single index and small scripts let multiple agents retrieve and add information about his work. Meyer says the system held 1,278 pages and achieved a 97.4% top-1 recall score on that date; the figures are his own measurements, not an independently verified benchmark.

Before you orderOffer from Amazon

Get everyday essentials delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

Meyer calls the system his Second Brain. Each memory is stored as a markdown page in a git repository, with each page covering a cluster of related facts, such as a project or decision. A file called INDEX.md lists the memories one line at a time. Meyer says the setup uses no vector database or cloud service, and that he can open and edit the files directly.

When an agent receives a question, a script called recall.py scores the index lines without opening all the memory pages. It opens the best-matching page and returns the section judged relevant, along with its source. Meyer says this lets him check which page informed an answer. A separate script, remember.py, writes a new page, its index entry and a log line together.

As of Nov. 21, Meyer reports 1,761 links between memories and 1,363 recalls. He defines top-1 recall as how often the first page selected is the right one and gives the system a score of 97.4%. The article does not describe the questions used, the scoring procedure or an external audit, so readers cannot assess from the account alone how the result would compare across other systems.

At a glance
reportWhen: Published Nov. 21, 2026; system metrics…
The developmentThorsten Meyer published a description of a shared, file-based memory system for his AI agents on Nov. 21, 2026.
AI memory · Briefing

What Is a Second Brain? How AI Agents Share Memory

Thorsten Meyer describes a personal knowledge base that lets his AI agents share and inspect long-term memory stored in markdown files and git.

97.4%
Author-reported top-1 recall
1,278
Memory pages on Nov. 21, 2026
1,761
Links between memories
1,363
Recalls reported

How shared memory works

Agents use one file-based store to look up and add information across sessions.

1
Catalogue memories

INDEX.md lists markdown pages about related facts, projects and decisions.

2
Find a relevant page

recall.py scores the index, then returns a relevant section with its source.

3
Save new information

remember.py writes a page, its index entry and a log line together.

What the design offers

The author’s setup emphasizes shared access and an inspectable record.

Inspectable

Read and edit files

The owner can open the memory directly and review its history in git.

Traceable

Follow the source

An answer points back to the page that informed it, helping the owner check the record.

Shared

One common memory

Multiple agents can look up information and add to the same store.

What remains unverified

The figures and design details come from Meyer’s account of his own system.

The account reports
  • A 97.4% top-1 recall score
  • Markdown files, an index and scripts
  • A setup intended for one owner
Open questions
  • Test questions and scoring method
  • Independent evaluation or comparisons
  • Team access, simultaneous edits and error handling
Read with careThe article does not provide an independent audit or enough testing detail to assess how the recall score compares with other systems.
The takeaway
A shared, inspectable file store can help AI agents carry information across sessions, but the reported recall score and wider usefulness still need independent evidence.

A Checkable Memory Across Agents

The system addresses a practical limitation of conversational AI: an agent may lack information from an earlier session unless that information is supplied again or saved somewhere it can access. A shared store can give several agents a common record of project details, past decisions and preferences, reducing repeated explanations in the author’s workflow.

Meyer’s design also makes the memory inspectable. Users can read and change text files, view their history in git and follow a returned answer to its cited page. Those features may help an owner spot stale or incorrect information. They do not, by themselves, establish that every answer is accurate or that the system can prevent an agent from misusing a memory.

The approach has trade-offs. It depends on maintaining useful index entries and organizing facts so the recall script can find them. Meyer describes a setup for one owner; the article does not explain access controls, conflict resolution or shared editing for a larger team.

Products Worth Considering

As an affiliate, we earn on qualifying purchases.

From Session Context to Stored Files

An AI agent typically uses information available in its current context window: the prompt, conversation and other material supplied during that session. Meyer says the problem he wanted to address was that details could be lost between sessions and that memory features in different tools were not always easy for him to inspect, edit or audit. The Second Brain is his answer to that personal workflow problem.

The account presents the system as a combination of plain files, an index and scripts, rather than as a general product launch. Git records changes to the files, while the index provides a compact catalogue for retrieval. The article and its accompanying explainer video describe the author’s implementation and include diagrams; they do not amount to a comparative study of AI memory methods.

“I got tired of being their memory. So I built them one.”

— Thorsten Meyer, describing his motivation

How the Recall Score Was Measured

The account does not specify the size or composition of the test set behind the 97.4% top-1 score, how a correct first result was determined, or whether the test included difficult and ambiguous queries. It also does not report an independent assessment or compare the score with another retrieval method. The figure should be read as the author’s measurement of his own system.

Other operational details remain open. Meyer describes a design intended for one owner but does not explain how it handles simultaneous edits, access permissions, sensitive information, backups or incorrect memories. The article also does not say how the system would perform with other agents, larger collections or different types of work.

More Views of the System

Meyer says the explainer video walks through nine views of the Second Brain, including a map of connections among projects, skills, routines and the apps that use the memory. The available account begins describing a daily brief but ends before explaining all nine views. It gives no date for a next release, independent evaluation or expansion to team use.

For now, the next step for readers interested in the method is to consult Meyer’s published explanation and video for details of the views and scripts. Any assessment of how well the approach generalizes will depend on further documentation, including how recall is tested and how the system manages access and errors.

Key Questions

What is Thorsten Meyer’s Second Brain?

It is a personal knowledge base that Meyer says he shares with his AI agents. Its memories are markdown files in git, catalogued in a single index.

How does an agent retrieve a memory?

According to Meyer, recall.py scores the index entries, opens the best-matching page and returns a relevant section with its source.

What does the 97.4% figure mean?

Meyer defines it as the rate at which the first page selected by the system is the right one. The account does not provide enough testing detail or independent verification to judge the score beyond his description.

Can this setup be used by a team?

The described setup is built for one owner. Meyer does not explain team permissions, simultaneous edits or how people would resolve conflicting changes.

Source: Thorsten Meyer AI

You May Also Like

When Does Cheap Memory Come Back? The 2027–2029 Question

Memory prices are expected to remain high through 2028–2029, with relief delayed beyond 2027 due to industry capacity constraints and demand trends.

Understanding Anthropic’s $965B Series H: The Compute Revolution

Anthropic’s latest funding round highlights a $965 billion valuation driven by a massive push to secure AI hardware infrastructure, including chips, memory, and power capacity.

7 AI-Driven Solutions That Will Dominant In 2026

Discover the seven AI-driven innovations expected to dominate in 2026, transforming industries and shaping the future of technology.

A Curated Look At 15 AI Image Generators For 2027 Creative Work

A 15-pick guide compares image-generation tools with prompt aids, specialized resources and technical learning material for creative work.