5 Ways A Second Brain Helps AI Agents Work Together
AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

Thorsten Meyer published a description of a shared “Second Brain” that lets his AI agents consult and update the same knowledge base. It uses Markdown files, a one-line-per-page index and scripts for recall and storage; Meyer reports 1,278 pages and a 97.4% top-one recall score as of Nov. 21, 2026.

Thorsten Meyer published a description of a shared memory system, which he calls a Second Brain, for the AI agents he runs. The setup gives those agents a common reference for his projects and decisions, addressing the need to repeat context when separate sessions begin without it.

Before you orderOffer from Amazon

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
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

Meyer says the system stores knowledge in plain Markdown files kept in Git, with each page covering a cluster of related facts. A file named INDEX.md lists each memory in one line. He says the setup uses no vector database or cloud service, and that the files can be opened and edited with ordinary text tools.

For recall, a script called recall.py scores the index lines, then opens the best-matching page and returns the section that answers the question, along with its source. Meyer says this lets him inspect where an agent’s answer came from. 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 system contained 1,278 pages and 1,761 links, and had handled 1,363 recalls. He reports a 97.4% top-one recall score, which he defines as how often the first page selected is the right one. These figures and the system’s performance are the author’s own account; the supplied material does not provide an independent evaluation.

At a glance
reportWhen: Published November 21, 2026
The developmentThorsten Meyer published an account of a shared, file-based memory system he uses across his AI agents.
AI Agents · Briefing

5 Ways a Second Brain Helps AI Agents Work Together

Thorsten Meyer describes a shared, readable memory system for the AI agents he runs. It aims to reduce repeated context and make answers easier to trace.

97.4%
reported top-one recall score
1,278
memory pages
1,761
links between pages
1,363
recalls handled

One memory, shared across agents

A common record can help agents resume work with project details and decisions already available.

Shared context

Less repeated setup

Agents can consult the same knowledge when sessions begin without prior context.

Readable files

Owner can inspect

Knowledge lives in Markdown files that can be opened and edited with ordinary text tools.

Traceable answers

Sources come back

Recall returns the relevant section along with its source page, helping the owner check answers.

How memory moves

Two scripts handle finding and adding information in the shared store.

1
Find a likely page

recall.py scores the one-line entries in INDEX.md.

2
Return sourced detail

The script opens the best match and returns the section that answers the question, with its source.

3
Store new knowledge

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

Built around simple files

The described setup keeps its knowledge in Git and uses a compact index to find pages.

The system includes
  • Markdown pages, each covering related facts
  • INDEX.md with one line per memory
  • Git history for file changes
Meyer says it avoids
  • A vector database
  • A cloud service
  • Memory that is hard for him to inspect
“One memory, shared by every agent I run, that I can open and read myself.”— Thorsten Meyer, describing the system

What readers should weigh

The account describes one owner’s setup; it does not establish performance across users or workloads.

  • The 97.4% score is Meyer’s report; test design and correctness criteria are not provided.
  • The setup is described as built for one owner, not a team.
  • Handling of sensitive, outdated or conflicting memories is not specified.
  • Useful pages and index entries still depend on owner maintenance.
The takeaway
A shared, inspectable file store could help agents carry project context forward, while its reported accuracy and wider team suitability remain unverified.

Shared Memory Across Agent Sessions

A common, readable memory could reduce repeated setup work when a person uses multiple agents or returns to a project in a new session. If agents consult the same record of decisions and project details, they can start with information that would otherwise need to be provided again.

The design also puts source checking and editing within reach: the knowledge lives in files the owner can inspect, and the recall process returns a source page. That could make errors easier to trace and correct. The arrangement depends on an owner maintaining useful pages and index entries, however, and Meyer describes it as a system built for one owner rather than a team.

Products Worth Considering

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite

Logitech MX Mechanical Mini for Mac Wireless Keyboard, Space Gray

Logitech MX Mechanical Mini for Mac Wireless Keyboard, Space Gray

As an affiliate, we earn on qualifying purchases.

Markdown Files and Index Led Recall

Meyer frames the problem as a limitation of agent sessions: an agent can use the information available in its current context, but that context does not necessarily carry over when the session ends. He says memory features can be difficult to read or audit and may be tied to individual tools. His system is his response to that problem, using one shared store across the agents he runs.

The described workflow has two parts. On recall, the script searches the compact index before opening a page. On storage, the write script adds the page, index line and log entry in one operation. Git records file history, according to Meyer. The source material also mentions nine system views and an explainer video, but does not fully describe all of those views.

““One memory, shared by every agent I run, that I can open and read myself.””

— Thorsten Meyer, describing the system

Independent Performance Checks Unreported

The reported 97.4% top-one score is not accompanied in the supplied material by details about how the test set was assembled, how correctness was judged or how often the system was evaluated. It is therefore unclear how the figure would compare with other memory systems or perform across different users and workloads.

The account also does not specify how sensitive information is protected, how conflicting or outdated memories are handled, or how the system would work when several people contribute. Meyer describes the design as intended for one owner; its suitability for shared organizational use is not established by the material.

Further Details on Use and Evaluation

Meyer’s post and explainer video provide the current description of the system. The supplied source does not announce a product release, an independent benchmark or a team deployment. Further information about the recall evaluation, maintenance process and handling of outdated or conflicting notes would help readers judge how well the approach works beyond the author’s own setup.

Key Questions

What is the Second Brain?

It is Meyer’s shared knowledge base for the AI agents he runs, made from Markdown files and an index that lists those files.

How does an agent find a stored fact?

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

What does the 97.4% figure measure?

Meyer defines it as the rate at which the first page selected is the right one. The supplied account does not detail an independent test or its methodology.

Can a team use the system?

Meyer says the setup is built for one owner, not a team. The source does not establish how it would handle multiple contributors.

Source: Thorsten Meyer AI

You May Also Like

How Google’s Pichai Sees AI Self-improvement In Its Early Stages

Google CEO Sundar Pichai says AI systems’ ability to improve themselves remains at an early stage; details and implications are not yet clear.

AI Video Editing Software In 2027: 12 Options For Different Creators

A buying guide compares 12 video-editing options, from guided Adobe apps to CyberLink and Nero, while flagging platform and license limits.

Alphabet plans to raise $80 billion from stock sales to fund AI buildout

Alphabet plans to raise $80 billion through stock offerings, including a $10 billion investment from Berkshire Hathaway, to fund its AI infrastructure growth.

Choose Boring Technology (2015)

Examining the 2015 concept of ‘Choose Boring Technology’ and its impact on innovation and industry practices.