Acorn Labs
// ACORN LABS

AI, made easy.

Just starting out, or already juggling several AIs? Pick your way in.

They draft. You send. I stay on the job.
grok · chat● Holdfast connected
> you: what did I quote Mrs Patel for the shelving?
check "quote Mrs Patel shelving"
NOT ON RECORD · no saved quote matches
> grok: That isn't on record, so I won't guess. Tell me the figure and I'll save it.
recall "rules"
Never send a message or email for me without my yes.
✓ rule sent before the first word
Example chat · check and recall are real Holdfast tools
// THE PROBLEM

Your AI knows the chat.
It doesn't know your business.

✕You type the same facts into every new chat.
✕ChatGPT doesn't see what Grok decided.
✕It quotes last month's price because nobody told it the new one.
✕You become the copy-paste cable between them.
ONE EXAMPLE: GROK BOTS

Each Grok bot remembers itself, not the company. Two bots on the same job don't share a record, and ChatGPT and Claude don't see what either of them did.

// WHAT I DO

Four things.

01
Memory.

Each chat only knows that chat. I give Grok, ChatGPT and Claude one shared record of your facts, rules and decisions.

02
Workers.

Different AIs get different jobs, working from the same rules and the same record.

03
Independent check.

Important work gets checked by a different model before you rely on it. Nothing goes out without your yes.

04
Stay on the job.

I keep it running, fix what drifts, and change it as the business changes.

// WHAT I CAN SET UP

Everything here I run in my own company first. I work remotely, wherever you are.

Grok, ChatGPT and Claude, set up for real work
Rules, shared context and your sign-off before anything goes out.
Ready-made or custom AI agents
Built for your business, working as a team, with rules and a check before anything goes out.
A second model checking the work
Anything that matters goes to a different AI to find mistakes first.
Shared memory across your AIs
Holdfast, the memory layer I built, so every AI starts from the same notes.
Connecting AI to your tools and files
Connectors so the AI can read what it needs, and only that.
Private AI on your own hardware
I review what you run now, then set up local models where your data has to stay in-house.
Automation that runs on a schedule
Reports, checks, backups and research that run without you remembering.
Research with sources
Bots that find and check information and file it where you can see it.
Websites and hosting
Fast, simple sites and the hosting behind them.
Cost control across models
The right model for each job, so you are not paying top rates for simple work.
Audit trails
A record of what each AI did and why, so you can check it later.
Your data, owned by you
Export it, back it up, delete it. Nothing locked in.
// WHY ME

Business first. Tools second.

Business first.

I start from how you work, not from the tools.

Same facts everywhere.

Every AI starts from the same notes and rules.

A second model checks the serious stuff.

That's a habit I run, not a Holdfast button.

You still send.

The AI drafts. Nothing leaves without you.

I install it.

You don't have to become the technical person.

I stay on the job.

Not a one-off setup that drifts.

// HOLDFAST

I built the bit I couldn't find.

Each chat only knows that chat. Holdfast is the shared record they come back to.

When I set you up, this is the memory layer I can install: one file of facts, rules and decisions that Grok, ChatGPT and Claude get when they connect.

It reminds them. It can't force them. The file stays on the machine running it, and your chats still go to the AI providers.

See how Holdfast works, and start a 7-day trial →
01 / 05Working for you
The AI problems it fixed, counted.
02 / 05What my AI knows
Every memory, who saved it, one click to undo.
03 / 05Add an AI
Claude Desktop in one click. ChatGPT and Grok with a link I set up.
04 / 05Setup wizard
Three steps. No terminal.
05 / 05Built-in help
Plain-English guides inside the app.
// INSIDE HOLDFAST
Real screens · hover to pause
// ACORN, MY OWN SETUP

I run my own company this way. Three things it taught me.

BRAIN :110 :111 :236
112 tests passed. The live path was still wrong.

Every test built the call by hand. The real program passed it a different way, so three safety guards did nothing.

BRAIN :42 :43
A working folder is not a sandbox.

A verifier's scratch script committed 58,000 files to my live code. Repaired the same morning; scratch jobs now carry a safety rule.

BRAIN :203 :204 :205
Making it capable is easier than deciding what it's allowed to do.

I built remote execution switched off and had DeepSeek attack it: PASS, 16 of 16. Then I armed it. It sits idle, and no job has been launched.

Read the full story →
// WHO IT'S FOR

You're already using AI for real work.
You're still holding the whole thing together.

// PRICES

Ask me what it costs to set this up.

Message me. I'll tell you if I can help and what that would cost.

With me
Setup and projects
quoted when you ask
Holdfast
Monthly, if I install the memory layer
ask me the price
7 days free, I set it up with you
Solo or team (up to 5)

Show me where AI is getting messy.

Chat with me directly. Tell me what you're trying to get AI to do, and I'll tell you honestly if I can help. I read and reply myself.

LINE@908zlitiMessengerAcorn LabsLinkedInChris RothwellX@thuntalai