Is an Obsidian or Notion vault enough for a second brain?

For writing and finding things, yes, and it is the right foundation. It stops when you need the system to act on what is in it. A file records that something was said. It cannot record that the thing was settled last Tuesday, and editing the note does not retract the summaries already built from it.

How technical is it to build one?

The vault itself can be simple. The moment you add email or chat it becomes a real engineering project: chunking, embeddings, a vector store, a scheduler, retries and a cost budget you have to watch. Stopping at the vault is a reasonable choice rather than a failure.

What does Team0 do that a prepared layer does not?

The World Model keeps individual facts rather than a regenerated summary, so each one can be retired when something settles it. Your Chief knows what is genuinely still open with each person, goes quiet on what got handled, and points at the email or meeting behind each claim. Correct a mistake and the World Model keeps that correction in force when old sources return. Then she acts on it: triages the inbox, drafts the replies, preps you before calls.

Why does my assistant keep telling me to do things I already did?

Because completion arrives through a different channel than the one that created the task, and a snapshot architecture never reconciles them. Fixing it means deciding which source of truth outranks which at the belief layer, not in the summary.

Comparison

Build your own second brain. Then meet month six.

A folder of markdown, linked up, searchable, and entirely yours is the right start. The hard part comes when email, meetings, and action begin changing what those files say. The Team0 World Model is built for that next problem: maintaining what is still true, retiring what is not, and keeping the source attached.

Free 7-day trial · No credit card

Side by side

Where each one actually wins.

Dimension	Your own build	Team0	Edge
Owning the files	Plain files on your disk. Portable, greppable, entirely yours.	Hosted, so there is nothing to run, tune or back up.	Your own build
Fit to how you work	A workflow you designed. Exact on the day you wrote it, and fixed until you rewrite it.	Learned from the work itself, as it happens. When the way you work changes, so does she.	Team0
Acting on what it knows	A vault is a place to read. Acting on it is another project entirely.	Triages the inbox, drafts the replies, preps you before calls, and carries a job over days on her own.	Team0
One person, four spellings	Forty notes, no merge. Search finds whichever spelling you typed.	Merged on exact keys like an email address. A near-match is only a suggestion until it is checked.	Team0
Knowing something went stale	A rebuilt summary has no memory of what it used to say.	The World Model keeps two dates on each fact: when it was true, and when Team0 learned it.	Team0
Fixing a mistake for good	Correct it, and the next run relearns it from the same file.	Correct or delete it, and the World Model keeps that correction in force when the source is read again.	Team0
Who maintains it	You, whenever the system needs another repair or rebuild.	Team0 maintains it as the work changes.	Team0

How it goes

01
Files are the right start.
Markdown in folders is portable, greppable, works offline and is independent of a vendor. If your second brain is a vault you write in daily, it is already doing something valuable.
The files are not the problem.
02
But a file is a document, and what you need is a fact.
"Dana needs the proposal by Thursday" is one sentence inside a note covering five other things. You cannot mark that sentence done, and editing the note later retracts nothing already built on top of it.
You can version a document. You cannot retire a paragraph.
03
Connect a real channel and the vault becomes a pipeline.
Email and chat are streams, not documents. Handling them means chunking, embeddings, a vector store, a scheduler, retries and an API key with a budget. Then the bill arrives, because raw history is being re-read on every run.
The vault was a weekend. The pipeline is a second job.
04
Precomputing fixes the bill and freezes the mistake.
Distilling once on a schedule into a prepared layer is the correct fix, and it removes the repetition you were actually paying for. But a snapshot cannot say "that is no longer true", so a closed matter reappears next cycle wearing fresh paint.
Anyone can store a fact. Retiring one is the whole discipline.
05
Which is the part we built.
The Team0 World Model maintains what is still open with each person and retires an item when evidence shows it was handled. It keeps the source attached so you can check it, preserves your corrections when old sources return, and says "amount not stated" rather than putting an invented number in front of you.
Not just memory. Maintained understanding.

The failure nobody warns you about

It will tell you to do something you already did.
Your prepared layer says the proposal is outstanding. The task is ticked in your tracker. You said in chat last week that you sent it. All three are in the system at once and the stale sentence wins, because it is the one that got summarised into the layer the agent actually reads. Nothing is broken. Every part did its job. The architecture simply never decided who outranks whom.

Questions

Is an Obsidian or Notion vault enough for a second brain?: For writing and finding things, yes, and it is the right foundation. It stops when you need the system to act on what is in it. A file records that something was said. It cannot record that the thing was settled last Tuesday, and editing the note does not retract the summaries already built from it.
How technical is it to build one?: The vault itself can be simple. The moment you add email or chat it becomes a real engineering project: chunking, embeddings, a vector store, a scheduler, retries and a cost budget you have to watch. Stopping at the vault is a reasonable choice rather than a failure.
What does Team0 do that a prepared layer does not?: The World Model keeps individual facts rather than a regenerated summary, so each one can be retired when something settles it. Your Chief knows what is genuinely still open with each person, goes quiet on what got handled, and points at the email or meeting behind each claim. Correct a mistake and the World Model keeps that correction in force when old sources return. Then she acts on it: triages the inbox, drafts the replies, preps you before calls.
Why does my assistant keep telling me to do things I already did?: Because completion arrives through a different channel than the one that created the task, and a snapshot architecture never reconciles them. Fixing it means deciding which source of truth outranks which at the belief layer, not in the summary.

Remembering is the easy half.

The Team0 World Model maintains what is still true, retires what is not, and keeps the source attached. That is the harder half.

Free 7-day trial · No credit card

See how the Team0 World Model works

Prepared by your Chief of Staff

REV 2026-08-12

Comparison

Build your own second brain. Then meet month six.

Free 7-day trial · No credit card

Side by side

Where each one actually wins.

Dimension	Your own build	Team0	Edge
Owning the files	Plain files on your disk. Portable, greppable, entirely yours.	Hosted, so there is nothing to run, tune or back up.	Your own build
Fit to how you work	A workflow you designed. Exact on the day you wrote it, and fixed until you rewrite it.	Learned from the work itself, as it happens. When the way you work changes, so does she.	Team0
Acting on what it knows	A vault is a place to read. Acting on it is another project entirely.	Triages the inbox, drafts the replies, preps you before calls, and carries a job over days on her own.	Team0
One person, four spellings	Forty notes, no merge. Search finds whichever spelling you typed.	Merged on exact keys like an email address. A near-match is only a suggestion until it is checked.	Team0
Knowing something went stale	A rebuilt summary has no memory of what it used to say.	The World Model keeps two dates on each fact: when it was true, and when Team0 learned it.	Team0
Fixing a mistake for good	Correct it, and the next run relearns it from the same file.	Correct or delete it, and the World Model keeps that correction in force when the source is read again.	Team0
Who maintains it	You, whenever the system needs another repair or rebuild.	Team0 maintains it as the work changes.	Team0

How it goes

01
Files are the right start.
Markdown in folders is portable, greppable, works offline and is independent of a vendor. If your second brain is a vault you write in daily, it is already doing something valuable.
The files are not the problem.
02
But a file is a document, and what you need is a fact.
"Dana needs the proposal by Thursday" is one sentence inside a note covering five other things. You cannot mark that sentence done, and editing the note later retracts nothing already built on top of it.
You can version a document. You cannot retire a paragraph.
03
Connect a real channel and the vault becomes a pipeline.
Email and chat are streams, not documents. Handling them means chunking, embeddings, a vector store, a scheduler, retries and an API key with a budget. Then the bill arrives, because raw history is being re-read on every run.
The vault was a weekend. The pipeline is a second job.
04
Precomputing fixes the bill and freezes the mistake.
Distilling once on a schedule into a prepared layer is the correct fix, and it removes the repetition you were actually paying for. But a snapshot cannot say "that is no longer true", so a closed matter reappears next cycle wearing fresh paint.
Anyone can store a fact. Retiring one is the whole discipline.
05
Which is the part we built.
The Team0 World Model maintains what is still open with each person and retires an item when evidence shows it was handled. It keeps the source attached so you can check it, preserves your corrections when old sources return, and says "amount not stated" rather than putting an invented number in front of you.
Not just memory. Maintained understanding.

The failure nobody warns you about

It will tell you to do something you already did.
Your prepared layer says the proposal is outstanding. The task is ticked in your tracker. You said in chat last week that you sent it. All three are in the system at once and the stale sentence wins, because it is the one that got summarised into the layer the agent actually reads. Nothing is broken. Every part did its job. The architecture simply never decided who outranks whom.

Questions

Is an Obsidian or Notion vault enough for a second brain?: For writing and finding things, yes, and it is the right foundation. It stops when you need the system to act on what is in it. A file records that something was said. It cannot record that the thing was settled last Tuesday, and editing the note does not retract the summaries already built from it.
How technical is it to build one?: The vault itself can be simple. The moment you add email or chat it becomes a real engineering project: chunking, embeddings, a vector store, a scheduler, retries and a cost budget you have to watch. Stopping at the vault is a reasonable choice rather than a failure.
What does Team0 do that a prepared layer does not?: The World Model keeps individual facts rather than a regenerated summary, so each one can be retired when something settles it. Your Chief knows what is genuinely still open with each person, goes quiet on what got handled, and points at the email or meeting behind each claim. Correct a mistake and the World Model keeps that correction in force when old sources return. Then she acts on it: triages the inbox, drafts the replies, preps you before calls.
Why does my assistant keep telling me to do things I already did?: Because completion arrives through a different channel than the one that created the task, and a snapshot architecture never reconciles them. Fixing it means deciding which source of truth outranks which at the belief layer, not in the summary.

Remembering is the easy half.

The Team0 World Model maintains what is still true, retires what is not, and keeps the source attached. That is the harder half.

Free 7-day trial · No credit card

See how the Team0 World Model works

Prepared by your Chief of Staff

REV 2026-08-12

Where each one actually wins.

Files are the right start.

But a file is a document, and what you need is a fact.

Connect a real channel and the vault becomes a pipeline.

Precomputing fixes the bill and freezes the mistake.

Which is the part we built.

Remembering is the easy half.

Where each one actually wins.

Files are the right start.

But a file is a document, and what you need is a fact.

Connect a real channel and the vault becomes a pipeline.

Precomputing fixes the bill and freezes the mistake.

Which is the part we built.

Remembering is the easy half.