AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: How My AI Agents Keep Context In One Shared Memory on ThorstenMeyerAI.com

Age 18–24?Offer from Amazon

Prime made for students and young adults

  • Fast, free delivery for dorm and study essentials
  • Prime Video and Amazon Music included
  • Member-only deals
Try Prime for Young Adults Free trial for eligible 18–24 year olds
As an affiliate, we earn on qualifying purchases.

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 and a Git-tracked index rather than a vector database or cloud service; Meyer reports 1,278 pages and a 97.4% top-1 recall score as of Nov. 21, 2026.

Thorsten Meyer says he built a shared, long-term memory for the AI agents he uses, allowing them to retrieve and add information across work sessions. In an article published Nov. 21, 2026, he describes a system of Markdown files stored in Git and a single index, intended to make agents’ remembered information readable and checkable by its owner.

Meyer calls the system his Second Brain. Each page stores a cluster of related facts, such as information about a project or a decision. A file named INDEX.md lists the pages, one line per memory. Meyer says agents consult that index before opening a relevant page, rather than searching every file.

A script called recall.py scores index entries, opens the page it ranks highest and returns the section relevant to the question, along with its source. Meyer says that source reference lets him inspect the underlying page when an agent uses the memory in an answer. Another script, remember.py, writes a new page, its index entry and a log entry together.

As of Nov. 21, 2026, Meyer reports that the system contained 1,278 pages and 1,761 links, and had recorded 1,363 recalls. He says its top-1 recall score — his measure of how often the first page selected is the right one — was 97.4%. These figures are reported by Meyer; his article does not provide an independent evaluation or the test method behind the score.

At a glance
announcementWhen: Published November 21, 2026; the system…
The developmentThorsten Meyer published a description of a shared, file-based memory system for the AI agents he uses.

A Memory Agents Can Share and Inspect

The design addresses a practical limitation of session-based agents: information available during one interaction may not carry into the next. A shared store can let different agents draw on the same project details and prior decisions, reducing the need for a user to repeat them. Meyer presents the ability to inspect the source page as a core feature, rather than relying on a memory store that users cannot readily review.

Using ordinary files also makes the setup easier to read and move than a system tied to a particular agent product, according to Meyer. Git provides a history of changes, while the combined write step is meant to keep the index in sync with the pages. Those choices may appeal to people who want control over stored information. They also put responsibility on the owner to maintain the files, scripts and access rules.

The model is not presented as a ready-made team service. Meyer says it is built for one owner, and identifies discipline as a cost. The article does not establish how well the approach scales to multiple users, how much work upkeep requires or whether its reported recall rate would hold in other settings.

Shark Pet Cordless Stick Vacuum, Up to 40-Minute Runtime, Grey, IX141

Shark Pet Cordless Stick Vacuum, Up to 40-Minute Runtime, Grey, IX141

Bose QuietComfort Headphones (2nd Gen), Wireless Noise-Cancelling Over-Ear Headphones, Black

Bose QuietComfort Headphones (2nd Gen), Wireless Noise-Cancelling Over-Ear Headphones, Black

Vitamix 5-Speed Immersion Blender, 625-Watt, Stainless Steel/Black

Vitamix 5-Speed Immersion Blender, 625-Watt, Stainless Steel/Black

As an affiliate, we earn on qualifying purchases.

From Session Context to Shared Files

An AI agent works with information available in its current context window. When a session ends, that working context may no longer be available, so users can find themselves repeating preferences, project facts or earlier decisions. Meyer says existing memory features did not meet his preference for a single store he could open, edit and audit across agents.

His alternative is deliberately file-based: plain text pages kept in Git, with one catalogue used for retrieval. The workflow has two parts. Agents use index-first recall to find a page, then return a relevant section with its source; new information goes through a single script that updates the page, index and log. Meyer also describes nine monitoring and presentation views, including a daily brief, a change chronicle, an explorer of page links and a console for services and machines.

The article describes Meyer’s own implementation, not a controlled comparison with other memory products. It says the system uses no vector database or cloud service. Meyer also says personal data in the explainer video and images has been protected, and that names shown are cover names or masked.

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

— Thorsten Meyer, describing his motivation

Performance and Team Use Remain Untested

The reported 97.4% top-1 score is not accompanied in the article by a description of the test set, scoring procedure or independent verification. It is therefore unclear how the result compares with other memory systems or how it would change across different questions and collections of notes. The article also does not give enough detail to assess error rates, privacy protections beyond the stated use of local files, or the time required to maintain the system.

Meyer says the setup is designed for one owner, leaving its suitability for shared organizational use unresolved. The source material also ends before completing its discussion of the system’s figures, so no further performance measures can be confirmed from the supplied account.

Testing the Workflow Beyond One Owner

Meyer’s article describes the system’s current structure and figures but does not announce a release, a product roadmap or a planned external evaluation. The next useful evidence would be a fuller account of how the recall score was calculated, what kinds of queries were tested and how often the top-ranked page failed to answer a question.

For readers considering a similar approach, the immediate next step is to review the described workflow against their own needs: whether a single index and plain files are manageable, who can edit them, and how changes are checked. Whether Meyer’s system will be adapted for teams or evaluated by others remains unclear.

Key Questions

What is Thorsten Meyer’s Second Brain?

It is a shared knowledge base for the AI agents Meyer uses. It consists of Markdown pages, a central index and scripts for retrieving and adding information.

How do the agents find a stored fact?

According to Meyer, a recall script first scores lines in the index, then opens the highest-ranked page and returns the relevant section with its source.

How large was the system on November 21, 2026?

Meyer reported 1,278 pages, 1,761 links and 1,363 recalls as of that date. He also reported a 97.4% top-1 recall score, without providing the testing method in the article.

Does the system use a vector database or cloud service?

Meyer says it uses neither: the memory is stored as plain Markdown files in Git, with a text index and local scripts.

Is it designed for teams?

Meyer describes it as a setup built for one owner. The article does not explain how it would handle multiple users, shared permissions or team maintenance.

Source: ThorstenMeyerAI.com

You May Also Like

Trade voice copilo

A new voice copilot tool is being tested for small trades businesses to streamline job notes and invoicing, promising time savings and improved cash flow.

What the Next Wave of Wearable Computing Looks Like

A glimpse into the next wave of wearable computing reveals seamless, immersive technology that will transform daily life—discover what’s coming next.

Are You Choosing A WiFi 7 Router For AI? Compare These 9

A buying guide compares nine WiFi 7 routers by band configuration, coverage, ports and ease of use; the best fit depends on your home and devices.

Synthetic Data: Training AI Without Real Samples

Imagine harnessing the power of AI without compromising privacy; discover how synthetic data revolutionizes model training and unlocks new possibilities.