What is an AI second brain? Judge every setup by who does the filing
By Matthew Firth, founder of ModelBrain · Oct 24, 2026 · 6 min readAn AI second brain is a store of your notes and decisions that an AI assistant can search and answer from, and most of them fail at the filing, not the searching. Here is our own proposal for judging one, the filing test: count what you must do by hand between having a thought and an assistant being able to recall it.
What an AI second brain is
Two pages that rank for this search define it in different ways. MindStudio (May 27, 2026) describes a persistent store that an AI agent queries by meaning, built from embeddings and a vector database. Tana (July 14, 2026) describes a system that fills, connects and updates itself from your work. The first definition is about retrieval. The second is about upkeep. Neither is wrong, but a tutorial that stops at retrieval skips the part most people give up on.
The filing test
A second brain has two jobs. It has to take things in, which means capturing a thought and putting it somewhere findable, and it has to give things back, which means retrieval. Retrieval is the part that has improved. The filing is still the part where setups succeed or fail, because it happens every day and a missed day leaves a gap that nothing fills.
So we score each setup on three questions: who captures a thought, who keeps it current when it changes, and what the assistant can do with it afterward.
| Setup | Who captures | Who keeps it current | What the assistant can do |
|---|---|---|---|
| A notes app you file yourself, with an AI on top | You write the note and put it in the right place | You review and update it. MindStudio expects people to curate content and review it from time to time for stale material | Search and answer from what you filed |
| A workspace that files for you (Tana, as its own page describes it) | The system captures from meetings, messages and documents, and files follow-up work as proposals you approve | The system updates a record when new information touches it, per Tana's page | Other tools such as Claude Code and Cursor can read and write through its MCP server, per Tana |
| A memory layer plus the notes you already have (ModelBrain) | Your assistant calls the remember tool during a chat. A folder of notes is read from where it already is | A newer memory can replace an older one. For a folder, the next sync replaces the old text with the new | Any connected assistant recalls from both, and answers point back to the note they came from |
What each setup costs you
The notes app keeps you in control and costs the most attention. It works well if you enjoy tending notes. It fails quietly on the days you do not, and the assistant on top can only answer from what you filed.
A workspace that files for you moves that attention onto approving what it proposes, and onto moving your work into its format. Everything we say about Tana comes from Tana's own marketing page.
A memory layer sits beside the notes you already have instead of replacing them. It does not remove filing. It removes the part where a decision made in a chat with an assistant has to be copied into a note by hand.
How ModelBrain handles your existing notes
ModelBrain reads folders of files you approve, and an Obsidian vault is just such a folder. There is no plugin, and Obsidian does not have to be running. Answers name the note they came from, so you can open it and check. When you edit a note, the next sync replaces the old text with the new, and ModelBrain does not keep earlier versions. When you delete one, it drops out on a later sync. Anything you leave out of the folder you add is never read.
The other half is memory that lives in no note. Your assistant saves it with the remember tool and finds it again with recall, in Claude Desktop, Claude Code, Cursor, or ChatGPT through a relay. For how this differs from Claude's own memory, see Claude memory explained, and for the Obsidian side, Does Obsidian have AI? and Obsidian MCP servers compared.
Where ModelBrain costs you something
ModelBrain is in beta and its engine is not open source today. The free local vault is not encrypted at rest. It does not yet notice when two of your notes contradict each other, which our fact sheet lists as not built. It reads folders and saves what your assistant remembers, and it does not pull in meetings, email or calendars on its own, which the workspace in row two does. ChatGPT needs a free account for the relay. Syncing the same memory across computers is paid, at $6 a device a month billed yearly.
Which to choose
If you like tending notes by hand, row one is enough and costs nothing extra. If you want the filing done for you and are willing to move your work into one product, row two is aimed at that. If you already keep notes in a folder, use more than one assistant, and want decisions from chats to land somewhere an assistant can find, row three adds the missing half.
Sources
- What is the AI second brain?, MindStudio, May 27, 2026
- What is an AI second brain?, Tana, July 14, 2026
- Use ModelBrain with an Obsidian vault, ModelBrain docs
- ModelBrain fact sheet
Common questions
What is an AI second brain? A store of your notes, decisions and documents that an AI assistant can search by meaning and answer from. Definitions differ on whether the system also captures and files things for you, which is the part to check before choosing one.
Do I need Obsidian to build an AI second brain? No. Any folder of text files, notes app or workspace can be the store. ModelBrain reads a folder you approve, and an Obsidian vault is one such folder, with no plugin needed.
Is an AI second brain the same as AI memory? They overlap. A second brain usually means your notes made searchable by an assistant. AI memory usually means facts an assistant saves while you work. ModelBrain covers both: it reads a notes folder and stores what your assistant remembers.
Last checked October 24, 2026. Settings and tool behavior change often, so confirm a detail on the vendor's own page before relying on it.