instibrain
Sign in
instibrain  /  use cases

Every team answers the same questions again.
Write them down once.

Six teams and the systems they run, one brain. For each: what goes in, whether our server can read it, and the answer an agent gives from it. Every answer written down today is read by every AI you hire from now on, so the brain is worth more each quarter than the one before.

one setting first

Every page carries a scope, and the scope decides who reads it: the whole team, one person, or only you. In the secure area the scope is also a key, held only by the devices you admit, so the server cannot read the page either.

sales

Quote with the real floor, not the one someone remembers.

The floor, the exception, and the deal you walked away from live in the heads of two people. Write them down once, and the draft is right before the call instead of corrected after it.

  • pricing/floors.mdsecure area
  • deals/declined.mdsecure area
  • playbook/objections.mdopen area
pricing/floors.mdsecure area
askedCan we go to $32k a year for Northwind?
agent, reading the pageNot without approval. The floor is $40k a year, and anything below it needs both founders to sign off.
Who can read itClaude Code on a laptop admitted to the finance scope. A keyless surface never sees this page.
support

The fix that worked last time, before the ticket escalates.

Someone already solved this once. When the answer is a page, the next person pays nothing for it, including the agent on their phone.

  • support/known-issues.mdopen area
  • runbooks/billing-sync.mdopen area
support/known-issues.mdopen area
askedA customer says invoices doubled after the sync ran. Seen it?
agent, reading the pageYes. It is the retry storm in known issues: drain the queue, then run the billing sync once by hand.
Who can read itAny agent you connect, Claude on the web or on a phone included.
engineering

Why the schema looks wrong, from the person who made it right.

The constraint that looks like a mistake is load-bearing. A decision record outlives whoever wrote it, and the agent reads it before it suggests the cleanup.

  • decisions/0012-soft-deletes.mdopen area
  • architecture/constraints.mdopen area
decisions/0012-soft-deletes.mdopen area
askedCan I drop the deleted_at column? Nothing reads it.
agent, reading the pageKeep it. Decision 0012: deletes must reach every device, so a row is marked, never removed.
Who can read itClaude Code in the repo, and any other agent you connect.
operations

The order things come back in, at three in the morning.

A runbook nobody can find is not a runbook. When it is a page, the person on call and their agent read the same steps, in the same order.

  • runbooks/restore.mdopen area
  • oncall/escalation.mdopen area
runbooks/restore.mdopen area
askedThe database is down. What first, and who do I wake?
agent, reading the pageBring back the database, then api, then web. If it is still down after twenty minutes, page the second name on the escalation list.
Who can read itAny agent you connect, including the one on the phone in your hand.
new hires

Day one is one command.

A new person runs one command: the sync client installs, the scope their role was granted lands as files, and their agent builds its own setup from pages in the brain. It arrives knowing the house style, the names, and where things live. A teammate's day one shows the whole of it.

  • onboarding/start-here.mdopen area
  • style/naming.mdopen area
style/naming.mdopen area
askedIs it a workspace or a project in our product?
agent, reading the pageNeither. The word is brain, and node for one page of it. Naming rules list the words we never use.
Who can read itEvery agent in the team scope, the new hire's included. A page like this is written to be read on day one by a stranger.
your systems

Your AI runs your systems. You never type a password.

The brain holds the map: which systems exist, who owns each one, where its credential lives, which command uses it, and what must never be done with it. The credential itself stays in your own vault; the agent composes the call and fetches the value at use time. instibrain never holds a secret.

  • access/map.mdsecure area
  • access/never.mdsecure area
access/map.mdsecure area
askedRotate the staging database password and update whatever uses it.
agent, reading the pageDone. The map names staging-db: owned by platform, credential in the vault at ops/staging-db, used by api and worker. Both restarted. Production is on the never list and was not touched.
Who can read itAn agent beside the sync client on a device admitted to the access scope. The vault holds the secret; the brain holds the map; our server holds ciphertext.
the fit

When a brain fits, and when it does not.

Four signs it fits your company, and three signs you want a different product.

it fits when

You hold facts you would not paste into a vendor's chat, and you want your AI to know them anyway. Someone on the team already runs an MCP client. Your knowledge is, or could be, markdown in a folder. You would rather own the memory than rent it with the model.

look elsewhere when

You want a chatbot over the raw sprawl: a brain is distilled and reviewed, not indexed. You want the vendor to run your agents: a brain stores them and never executes one. You want the model bundled: you bring the AI you already pay for.

Start with one team's folder.

Point the sync client at the markdown you already have, and connect the agents you already pay for.

See day one → Read the FAQ