Small businesses are building their own AI command centre. Here's what it actually costs.

A small-business owner rang me a few months back, genuinely frustrated. He said something I've not been able to shake since. "I've got Xero, a job-management system, three email accounts, and a Slack workspace. Every morning I spend the better part of an hour working out what I actually need to pay attention to. My enterprise clients have dashboards for all this. I've just got tabs."

I didn't have a good answer for him. I made some noises about reporting integrations and custom dashboards, the sort of thing that takes weeks and costs real money. We left it there.

What bothers me, looking back, is that the answer was already out there. I just hadn't been paying attention. A whole crowd of solo founders and small-business owners had quietly been building their own version of his enterprise dashboard, using free tools, over a weekend. They weren't waiting for a vendor. They weren't paying the tens of thousands a year that a Looker-style deployment can cost. They were wiring up an AI to all the systems they already had and asking it to tell them what mattered each morning.

I've been doing web and software work for twenty-odd years. It's a slightly humbling thing to discover that the most interesting development in small-business tooling came from people building in public on the internet, not from the enterprise software I spend my days around. The cobbler's children, and all that.

So I went and tested it properly. Here's what I found, including the parts the people selling these setups would rather skip past.

What an "AI command centre" actually means

Let's strip the term down, because it sounds grander than it is. An AI command centre isn't a product you buy. It's a small stack of free, self-hosted tools that does four things:

1. Connects to the business data you already have (email, calendar, accounting, documents, a CRM)
2. Answers questions about that data in plain English ("what's my outstanding invoice total right now?")
3. Sends you proactive summaries on a schedule (a 7am morning brief, a Friday-afternoon weekly review)
4. Acts on your behalf for simple recurring tasks (drafting an email, flagging something urgent)

The thing that makes this different from just opening ChatGPT is that these tools remember context, connect to your live data, and can act, not just chat. The difference is roughly the difference between asking a clever stranger for directions and having someone in the passenger seat who knows your car, your route, and where you're trying to get to.

That's the genuinely useful framing. It's also where I want to be careful, because there's a lot of noise around this right now. Half the posts about these tools are written by people selling weekend-setup packages, and the numbers they quote are marketing, not measurement. I'll come back to that. First, the tools.

The three tools worth knowing about

There are dozens of these things. Three are doing most of the actual work in small businesses, so those are the three I'll cover. All three are free and open-source.

OpenClaw: the "AI chief of staff"

OpenClaw is the one everyone's talking about. It's an AI agent that runs on your own machine (or a cheap server) and talks to you through apps you already use, like WhatsApp, Telegram, or Slack. No new interface to learn. You message it the way you'd message a colleague, and it can read your email, manage your calendar, browse the web, run files, and execute custom automations.

The reason it's caught fire is that it's proactive. It doesn't just answer when asked. Here's one of the posts that got people's attention, and it's the one that made me sit up:

Lacy

@lacybuilds

OpenClaw is doing my business taxes.

Without any prompting, it EMAILED my registered agent to verify that the annual report had been filed.

This is the most amazing thing I've ever seen. I officially can't live without it.

0

16 June 2026

A bot that, without being asked, emails your registered agent to confirm an annual report was filed. That's the dream and the warning sign in a single tweet. It's genuinely clever. It's also an AI sending email on your behalf to a third party about your legal compliance, which is a lot of trust to hand over. Sit with that for a second, because we'll need it later.

People are building real things with it. One developer set up a bot that finds local businesses without a website, builds them a preview, and emails the link automatically, every day, on autopilot (@techxsarfraj, 2026). Whether that's a good outreach strategy or just automated spam is a separate argument. The point is the capability is real and it's running unattended.

Setup is moderate difficulty. There's a command-line install and you'll need an API key. If the phrase "command-line install" just made you tense up, that's a useful signal about whether you should be doing this yourself.

Open WebUI: a private ChatGPT you control

Open WebUI is the most approachable of the three. It's a browser-based ChatGPT clone that runs on your own server or computer. You can point it at cloud models like Claude or GPT, or run a local model so nothing leaves your building at all. It handles multiple users, permissions, and shared history, so a small team can use one private instance instead of everyone pasting client data into a public chatbot.

This is the cost-saving story in one tidy package, and someone framed it better than I could:

Shruti Codes

@Shruti_0810

People are spending hundreds every month on AI tools.

A small box on a desk is quietly breaking that business model.

Mac Mini.
Ollama.
Open WebUI.

Unlimited AI.
One-time cost.

19

6 June 2026

"A small box on a desk." That's a Mac Mini running Ollama and Open WebUI, and for a lot of small businesses it really is a one-time hardware cost instead of a stack of monthly AI subscriptions. I like this honesty more than the OpenClaw hype, because it names the actual trade. You pay for hardware once, you do the setup, and in exchange you stop renting.

For an eight-person trades business worried about pasting client details into someone else's servers, this is the obvious starting point. Load it with your pricing guide and procedures, and new staff get answers from your own knowledge base. Setup is genuinely easy, usually a single Docker command.

AnythingLLM: for businesses drowning in documents

AnythingLLM is built around making your documents queryable. You feed it PDFs, Word files, spreadsheets, even web pages, and then you chat with all of it at once. You can set up separate workspaces per department, run it on local models for privacy, and track usage across a team. The installer is a proper graphical one, so no terminal required.

The use case that sold me was watching a partner at a small law firm ask it to compare two versions of a supplier contract and flag the differences. It took about 40 seconds. Her alternative was reading both documents herself, which is exactly the sort of three-hour job that AI should be eating. For law, accounting, property, or consulting, where the work is mostly documents, this is the most immediately useful of the three.

How they compare

What you actually get to see

The abstract idea of "visibility" doesn't land until you picture it against your own business, so here's what it looks like in practice.

A morning brief might be: every weekday at 7am, send me overnight revenue from Xero, today's appointments, any emails marked urgent, and any jobs due for completion today. A weekly review might be: every Friday at 4pm, message me total revenue versus last week, my top three clients by jobs completed, any invoices over 30 days outstanding, and team hours by person.

Then there are the questions you can ask whenever you want. What's my average job-completion time this month? Which clients haven't heard from us in 60 days? What are my payment terms with that supplier? And the proactive alerts: if any email contains the word "complaint" or "refund", message me straight away.

I know this sounds exactly like what enterprise software promised business owners a decade ago and mostly failed to deliver. The honest difference is that this version works, it costs close to nothing, and you're not waiting on a vendor's roadmap. (I'm aware that's a low bar. Enterprise software set it.)

The catch nobody selling this wants to discuss

Here's where I have to be the wet blanket, because the gap between a viral demo and a tool you'd run your business on is wider than the tweets suggest.

Start with the numbers. You'll see claims like "get 12 to 16 hours back per month" and "a part-time hire's worth of capacity for $20" (@MoneyKrabs123, 2026). Treat those as illustrative, not measured. They come from people whose business is selling you the setup. The real savings depend entirely on your workflow, and the real cost is never $20, because "free software" still costs your time, and time is the one thing small-business owners actually run short of.

Then there's security, and this is the big one. An AI command centre is, by design, an AI agent with access to your real accounts. Your email. Your calendar. Sometimes your accounting. Remember that tweet about the bot emailing a registered agent unprompted? Charming when it works. Now picture it misreading something and emailing the wrong party, or a misconfigured agent with email access getting compromised. You've handed account-level access to software you installed off the internet on a weekend. That's not a reason to avoid it. It's a reason to set it up properly, scope permissions tightly, keep it on infrastructure you control, and never paste live credentials into a shared cloud model.

Here's what genuinely worries me: thousands of non-technical people are wiring AI into their live business accounts after watching a YouTube video. Sooner or later that produces a notable security incident, a credential leak or a data exposure that makes the news. Not because the tools are bad, but because the maths on that many people granting that much access is unforgiving.

Last, the question I always come back to with self-hosted anything: who fixes it when it breaks? There's no vendor support line. No SLA. When OpenClaw pushes an update that breaks your morning brief at 6:45am, or your local model starts giving odd answers, you own that. For a developer who enjoys tinkering, that's a Saturday. For a café owner, that's the morning brief silently stopped working three weeks ago and nobody noticed.

If you want the other side of the local-AI privacy argument in more depth, I've written about it here:

Related Article11 min read

Running AI Locally: Why Australian Businesses Are Ditching Cloud AI for Total Privacy

77% of employees paste sensitive data into ChatGPT. With Australia's December 2026 disclosure deadline looming, local AI deployment isn't paranoia....

Read full article

And if you want the cautionary tale of how fast one of these tools went from indie project to trademark fight, this one's worth your time:

Related Article9 min read

The $600 Mac Mini That Runs Your Business 24/7: Inside the CLAWD Phenomenon

Mac Minis are reportedly selling out for CLAWD. We investigated if Australian businesses should care, and whether you actually need that $600...

Read full article

Where Webcoda comes in, honestly

I'm not going to pretend you need us to get started. You don't. Go and install Open WebUI this afternoon. It'll take you twenty minutes to have something working, and you'll learn more in that twenty minutes than in this whole article. For a solo operator who's comfortable with a bit of setup, the DIY path is genuinely fine.

The point where it stops being a weekend hobby and starts being a real business system is the point where we're useful. That's the difference between a chatbot loaded with a few PDFs and an AI that's safely connected to your live Xero, MYOB, HubSpot, Umbraco, or SharePoint data, sending the right people the right reports automatically, without exposing anything sensitive.

Concretely, what we'd actually do for an Australian SMB is this. We scope your data sources and work out which tool fits (not always the trendy one). We build the integrations to your real systems, including the custom ones that don't exist off the shelf, like pulling this week's completed jobs from your job-management software and summarising them. We lock down the security so client and financial data stays on infrastructure you control rather than leaking into a public model. We set up the three to five automations your business actually needs, train your team, and hand over a documented system you understand and own. You pay once for the setup. The software stays free.

That last part matters to me. The thing I dislike most about enterprise BI tools isn't the price. It's that you never really own them. This is the opposite. We build it, document it, and leave you with something that's yours.

Back to the bloke with the tabs

I rang my frustrated client back after I'd tested all this. Showed him what a morning brief could look like for his business: revenue overnight, today's jobs, the two emails that genuinely needed him before lunch, all in one message on the phone he already carries. He looked at it for a second and said, "Why did nobody tell me I could do this?"

The honest answer is that I didn't know either, until recently. Nobody fully has the measure of this yet. The tools are weeks-old in some cases, the security practices are still being worked out, and half the advice online is written by people with something to sell. We're all feeling our way through it together, me included.

What I'm confident about is the shape of the thing. The visibility small businesses have been told for years they couldn't afford is now genuinely within reach, for the cost of some hardware and a careful weekend, or one conversation with us if you'd rather not own the breakage yourself.

If you want to see how AI currently sees your business online, our AI Accessibility Checker is a sensible first step. But the more useful conversation is the one where you tell us your biggest visibility problem, the thing you spend 45 minutes every morning squinting at, and we work out whether a command centre actually solves it. Sometimes it does. Sometimes it doesn't, and I'll tell you that too.