Case study
The second brain I actually use
Not a demo. A system switched on and running against my real work — my clients, my voice notes, my thinking — every day for months.
Kevin Factor · Cape Town · Built on GoBot, the open-source agent by Goda Go / Autonomee. The voice pipeline, review gate, knowledge sync and client deployment are my own layer on top.
You are not short of information. You are drowning in it.
Every good idea you had this month exists somewhere — a voice note, a meeting, a message you sent yourself at 11pm. None of it is anywhere you can reach when it matters. The tools did not fail. They were never built to carry your context, only to hold your files.
A chat window is a genius with no memory, no hands, and no reason to care.
A brain is useless until you give it hands.
Claude on its own can reason but cannot touch anything. Connect it to a real service — my calendar, my notes, my client folders — and “did Sarah reply?” stops being something it guesses at and becomes something it checks.
Everything before that is a chatbot. Everything after it is an operator.
The layer that mattered
I talk. It files. Nothing gets lost again.
I think out loud in the car. Those recordings used to die in a folder — meeting tools only capture live calls and will not touch a recording you already made.
Raw words saved first
The transcript is stored before anything interprets it, so if the interpretation is wrong the original survives.
Nothing passes itself
Structured notes wait at a review gate until I approve them. Nothing reaches the searchable brain on its own.
A month of thinking out loud becomes a month of thinking I can search. That is the payoff — not the transcription.
It remembers by meaning, not by keyword.
“That thing about the pricing model” finds the right note, even though I never used those words. It grades its own recall against a fixed set of questions and fails the build if it gets worse.
- 13
- memory tables
- 4
- search by meaning, not keyword
- 8
- specialist agents
- 9
- jobs running in the background
Eight agents, each with a different way of thinking — research, content, finance, strategy, and a critic whose whole job is finding the hole in the plan. One command puts all of them on the same question and hands me the disagreement.
Why I trust it with real work
You cannot outsmart a trap. You can remove the exit.
Any assistant holding private data, reading untrusted text, and able to reach the internet can in principle be tricked into leaking. Most products promise to detect the trick. I treat that as unwinnable, so the design removes the exit instead.
- 01Nothing goes out by default. Outbound is denied unless the destination is verifiably mine.
- 02A secret cannot be spelled out slowly. Fetches are capped per domain, so drip-feeding across subdomains collapses onto one counter.
- 03Widening the gate happens off-channel. Permission is time-boxed and granted somewhere the model cannot reach.
- 04If the guard breaks, it locks. A corrupted state file resets to full restriction, never to open.
And if a conversation touched untrusted text, deletions from memory are held back and shown to me instead of applied.
Same conversation. Entirely different relationship.
| A chat window | This | |
|---|---|---|
| Memory | Starts empty, every time | Knows what you said in March |
| Reach | Describes what it would do | Opens the file and does it |
| Initiative | Waits to be asked | Briefs you before you ask |
| Voice notes | Stay in the folder | Become searchable thinking |
| Where it lives | Someone else’s product | Your machine, your rules |
| Getting better | When they ship an update | Every week you use it |
The gap is not intelligence — it is the same model underneath. The gap is context: what it is allowed to remember, and what it is allowed to touch.
You do not build this. You switch it on, then grow it.
Day one is a chat that answers. What turns it into a second brain is the order you add things in — and that order is the part I teach. Most people try to build the whole thing at once and abandon it in week two.
- 01
It works
A chat that replies, on your phone, today.
- 02
It is always there
Awake when you are not.
- 03
It thinks with you
Memory, and your voice notes flowing in.
- 04
It is properly yours
Your files, your clients, your guardrails.
The tool is free and open — the sequence is what saves you the six months.
Bring me the problem.
This is what the approach looks like when the client is me. Yours would be built around your work, not mine.
Start a conversation