Can AI Agents Share One Memory? Meet The Second Brain
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Can AI Agents Share One Memory? Meet The Second Brain on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get hardware and tech 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.

TL;DR

Thorsten Meyer says he built a personal knowledge base that lets his AI agents share long-term memory across sessions. It uses Markdown files, Git and a searchable 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. The system is designed for one owner, and its performance figures have not been independently verified.

Thorsten Meyer has described a shared memory system that lets his AI agents retrieve and add information across separate sessions, addressing a common limitation: agents generally lose the working context of a session when it ends. In an article published Nov. 21, Meyer said his “Second Brain” contains 1,278 pages and recorded 1,363 recalls as of that date; the figures and reported accuracy are his own, not independently audited.

The system stores information in plain Markdown files tracked with Git. Each page holds a cluster of related facts, such as information about a project or a decision. A single file, INDEX.md, lists the pages in one-line entries. Meyer says agents consult that index first, rather than reading every file or searching a vector database.

According to Meyer, a script called recall.py scores index entries, opens the page it judges most relevant and returns the section that answers the query, along with its source. The approach is intended to make an agent’s reference traceable: a user can inspect the cited file and correct it. A second script, remember.py, writes a new page, its index entry and a log entry together, reducing the chance that the catalogue and the underlying notes become inconsistent.

Meyer reports that the collection had 1,761 links and had handled 1,363 recalls by Nov. 21. He also gives a 97.4% top-1 score, which he defines as how often the first page selected is the right one. The article does not provide an evaluation protocol, test set or independent assessment for that figure, so it should be read as a measure reported by the system’s creator.

At a glance
reportWhen: Published Nov. 21, 2026; system figures…
The developmentThorsten Meyer published a description of a shared, file-based memory system for his AI agents, reporting its design and usage statistics as of Nov. 21, 2026.
Crypto market snapshot
Fear & Greed Index
12/100 — Extreme Fear
Bitcoin BTC$60,401▼ 2.4%
Ethereum ETH$1,698▼ 3.0%
Tether USDT$0.9988▼ 0.0%
BNB BNB$547.91▼ 3.8%
USDC USDC$0.9994▼ 0.0%
XRP XRP$0.9267▼ 2.4%
Solana SOL$70.49▼ 1.1%
TRON TRX$0.3065▲ 0.3%
Live data · CoinGecko · alternative.me (24h change)

A Shared Record Across Agent Sessions

A common memory could reduce the need for users to repeat project details, prior decisions and failed approaches every time they start a new agent session. It may also help when a person uses several agents or tools: the knowledge is kept in one place rather than divided among separate product-specific memory features.

Meyer’s choice of readable, version-controlled files emphasizes inspectability and portability. Users can open the notes with ordinary software, review edits in Git and trace an answer to a source page. Those features address concerns about hidden or difficult-to-audit memory, though they do not by themselves establish that retrieved information is accurate or that every agent will use it appropriately.

The design also has limits. Meyer describes it as a system for one owner, not a shared team workspace. A company adopting a similar setup would need to address permissions, privacy, access control and the process for resolving conflicting or outdated information. The source does not describe a formal security review or team deployment.

Products Worth Considering

LG UltraGear 34-Inch Curved QHD Gaming Monitor, 160Hz, FreeSync Premium

LG UltraGear 34-Inch Curved QHD Gaming Monitor, 160Hz, FreeSync Premium

Ledger Nano X Crypto Hardware Wallet – Bluetooth (Black + Billfodl)

Ledger Nano X Crypto Hardware Wallet – Bluetooth (Black + Billfodl)

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

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

As an affiliate, we earn on qualifying purchases.

From Session Context to Stored Notes

AI agents generate responses using information available to them in a current interaction, including the contents of their context window. Meyer’s article says that when a session ends, that session’s context is gone, leaving the user to repeat relevant background in a later conversation. Some software offers memory features, but Meyer says he wanted a single record he could read and edit across the agents he uses.

His proposed answer is a folder of text files and an index rather than a hosted memory service. Git preserves revision history, while the index gives the recall script a compact catalogue to search. The system is therefore not presented as a new AI model or a general-purpose commercial product; it is Meyer’s account of a workflow and its supporting scripts.

The article also describes nine views for monitoring and organizing the collection, including a map of connections, a daily brief, a triage process for session notes, a change log and a live view of agent activity. Meyer says a triage step determines whether new captures become permanent memories, are merged into existing pages or are discarded.

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

— Thorsten Meyer, describing the Second Brain

Accuracy, Privacy and Team Use

The article does not explain how Meyer calculated the 97.4% top-1 score, how many queries were tested or what counted as a correct page. It also does not compare the system with other retrieval methods. The reported figure is a creator-provided metric, not evidence of performance across other users, topics or agent platforms.

Other implementation details remain open. The source does not specify how access to personal files is restricted, how sensitive information is protected, how stale or contradictory memories are flagged, or how the scripts behave when the index selects the wrong page. Meyer says personal information in the accompanying video and frames is masked, but that does not amount to a broader security assessment.

It is also unclear whether Meyer plans to publish the code, package the system for other users or support multi-user collaboration. His article characterizes the setup as built for one owner, so wider use would require decisions about permissions and responsibility for shared edits.

More Detail on the Memory Workflow

Meyer’s article points readers to an explainer video showing the system’s views and its recall-and-write process. The next practical step for readers is to review those materials and the underlying scripts, if available, before treating the reported setup as reproducible or suitable for sensitive information.

The source does not announce a launch date, public release or independent evaluation. Further reporting would need to establish whether the code and evaluation method will be made available, and whether the system can preserve its traceability and reliability when used by multiple people or with different agent tools.

Key Questions

What is the Second Brain?

It is Thorsten Meyer’s personal knowledge base, which he says his AI agents can use to retrieve and add information across sessions.

How does an agent find information in it?

Meyer says the agent checks a one-line-per-page index, then a script selects a relevant file and returns an answering section with its source.

What does the 97.4% figure measure?

Meyer describes it as the system’s top-1 recall score: how often the first page selected is the right one. The article does not provide a test method or independent verification.

Can teams use the system?

The described setup is designed for one owner. The article does not establish team features, access controls or a shared editing model.

Does it use a vector database or cloud service?

Meyer says the system uses Markdown files tracked in Git and a local index, with no vector database or cloud service in the described setup.

Source: ThorstenMeyerAI.com

Nothing in this article is financial or investment advice. Cryptocurrency and precious-metal investments carry significant risk — do your own research and consider a licensed advisor.
You May Also Like

Why These Thunderbolt Docking Stations Are Perfect For AI In 2026

Explore why Thunderbolt docking stations like Dell SD25TB4, Anker TB5, and Plugable are ideal for AI workflows in 2026, supporting high-speed data, multiple displays, and power needs.

Breaking Down OpenAI’s Jalapeño Chip: Can It Truly Lead AI Innovation?

OpenAI releases initial performance results for its Jalapeño inference chip, showing promising efficiency gains over NVIDIA, but deployment and independent validation remain pending.

What Does SoftBank’s Quarterly Loss Reveal About AI’s Power Needs?

SoftBank reported a ¥1.12 trillion quarterly loss, including charges tied to power equipment bought for U.S. AI data center plans.

Runway Opens Praxis-1 As OpenAI Rethinks Its Robotics Team

Runway released open-weight robot model Praxis-1 as OpenAI shifted its robotics teams into broader research and applied groups.