A second brain that runs the agency.
Loopsuit Brain turns 477 expert videos into linked notes. Agents read those notes to price deals, draft proposals and plan the week, and every answer points back to its source.
Quote $7,000 flat for a 4 to 6 week scoped build, not hourly. Anchor it at 10 to 30% of the yearly value, and take the money upfront.
Better answers come from reading less.
The whole idea in 60 seconds, one product decision at a time.
Hours of video. Zero recall.
We learn from long creator videos, then forget them by Friday. So every transcript gets split into small linked pages, each one ending with its source.
guarantee them a certain number of meetings in a certain period of time and if you don't hit that, don't charge them at all. take the money upfront so they're locked in. and price the thing you deliver, never your hours. charge ten to thirty percent of what the problem costs them every year…
X meetings in Y days, or they don't pay.
Price the result, not the hours.
Charge 10 to 30% of the yearly value.
One brain, six subjects.
Each subject lives in its own vault with its own index, so an agent only reads what the task needs.
Counted with find, not from memory.
43,155 wikilinks
Pages point at each other, so one question pulls in the pages around it.
477 videos cited
Every page ends with a Source heading, so coverage is counted by a script.
The four C's
Plain markdown and git
Opens in Obsidian, Claude Code or Codex. No lock-in, full history.
Read the right vault, skip the rest.
A root router file tells any agent where things live. It opens one vault's index, follows links, and leaves the other five closed.
Read the vault's index first, then follow links.
Knowledge that does the work.
Seven skills turn the notes into agency output: a priced quote, a proposal, a cold email, a discovery script.
The brain doesn't take an agent's word for it.
Agents over-report. So every ingest run follows four rules, and a script checks the vault after every batch.
Every page ends with ## Source and the video link, so a script can count coverage.
Page counts come from find and grep on disk, never from the agent's summary.
rebuild-wiki.py --check runs after each batch, not once at the end.
One idea per page, linked in. Deepen a page before adding a new one.
"Report only counts you measured. Agents have inflated their output about 2x here, repeatedly."From the router file every agent reads first.
Now a product: Cortex.
The same loop, packaged for other people: capture a thought, let agents clean and link it, then explore the graph. Core loop built, billing next.
Want a brain like this for your team?
We turn your calls, docs and playbooks into a linked knowledge base your agents can actually use. Start with a free 20-minute call.