I set up an AI agent with its own computer and its own logins. Your team messages it the way they'd message anyone else, and it gets on with the work across the tools you connect.
It gets better at the jobs you give it, and it runs on whichever AI model you choose.
Get in touchIt has its own computer, its own logins and a memory of how your business runs.
It runs on a computer of its own. Brief it at 9pm and the answer is waiting in the morning. A job set for 6am Monday runs at 6am Monday, whoever is in.
It lives in Slack, Teams, Telegram or another tool we connect, so anyone you've given access can brief it from their phone, one to one or in a group chat. They're all talking to the same agent, with the same memory.
It remembers your customers, your projects and the way you like things done, and a method that worked can be saved as a skill and used again. The tenth invoice run needs far less explaining than the first.
A Monday brief on what's overdue. A heads-up when a quote has gone a week without a reply. If nothing needs you, it stays quiet.
It has a browser and logins of its own, so it can work through the supplier portal that connects to nothing, or the booking system nobody has updated since 2011.
It runs on a computer or server you control and works with Claude, GPT, Gemini or open models. Cheaper models can handle background tasks, and you can switch the main one as better options come out. For work that needs local processing, I can scope a model on your own hardware and show you which connected services still receive data.
An AI agent can be one step inside a workflow or a feature inside another product. This is a separate setup: it stays available, works across the tools you connect and takes different briefs over time.
A lead arrives, it drafts the reply, and the draft lands in the CRM for review.
One job, wired up once. The trigger, the steps and the result are settled in advance, so it runs the same way every time without anyone asking. Quicker to get live, and for most jobs it's the right answer.
More on the simpler option“Pull the outstanding jobs, summarise what changed and prepare the follow-ups.”
One brief can cover a range of related jobs. You ask in plain language, it chooses from the systems you have connected and decides the order, then comes back with the result or stops to ask. Predictable parts can still run as workflows; the agent handles the judgement and exceptions.
See how far it goesThe work comes back regularly, crosses several systems and changes course depending on what it finds.
One setup can cover a range of related jobs without building a new workflow for each one. If the same steps work every time, a workflow is usually the better answer.
Its limits come from the tools, access and approval rules you give it.
You ask for one thing
“Check which overdue invoices need attention. Leave out anyone who has already confirmed when they will pay.”
Back to you
A few drafts to check, and the one dispute that needs you.
Examples. Yours would run on whatever you already use.
I use OpenClaw and Hermes for real work. Each runs on its own computer.
Both have similar access to my tools and projects, including a browser for checking sites and working in web apps. I decide which projects each can see and which actions need my review.
I hand them jobs the way I'd hand them to someone on the team. Add a section to a site, make the change and bring it back for me to review before it goes live.
OpenClaw and Hermes are both open source and free to use, so the setup stays yours.
Comes with a large public library of skills other people have built and shared, so a common job often has a starting point already. Its instructions, memory and skills are plain files you can open and read, so you can always see what it has been told.
It teaches itself. When it works out how to do something, it writes the method down as a skill without being asked and uses it next time, so it gets quicker at your jobs the more you use it. For a bigger job, it can split the work between helpers running side by side.
They cover much the same ground. I'll recommend one once I know the jobs, the tools you use and where the agent needs to run.
The agent doesn't arrive knowing your business. Everything below gets decided with you, tested on real work and written down before I hand it over.
We start with a couple of jobs and only the access those jobs need. You decide what it can do on its own and where it stops to ask.
As it takes on more, its access grows with it, one job and one connection at a time.
If one predictable set of steps covers the job, an autonomous agent is more machinery than you need.
I'll tell you that and point you at a workflow instead.
More about AI workflowsYour agent installed, secured and connected, ready for you to start handing it work.
Your agent doing the first jobs you want to hand it, working and tested before you take over.
Everything in the core setup, plus:
All prices are in AUD, plus GST. Running costs, including hosting and model usage, are billed to you directly. There's no monthly fee from me; ongoing care is optional and scoped separately. Your own hardware and unusual integrations are quoted separately.
Want more out of the assistant you already use before you build a separate one? The smaller setup session is $750. See the assistant setup
OpenClaw and Hermes are free to use. You may still pay for the machine or hosting, the AI model and any paid tools the agent uses. Those bills come straight to you. I set out the likely running costs before we start.
On a computer or server we choose together. I recommend a separate machine or cloud server, rather than the computer you use every day. That makes its access easier to contain and keeps it available when your own computer is off. I'll quote a setup on your own hardware separately.
The agent stores its conversation history on the computer or server where it runs. If it uses Claude, GPT or another hosted model, the messages needed for the job are sent to that provider. A fully local setup is possible, but it usually means more capable hardware or a smaller model.
Yes. We can add jobs and connections as you find more useful work for it. Before anything new goes live, I check what extra access it needs and where it should stop for approval. Small additions may fit into ongoing care; larger ones get their own quote.
No. Core setup includes a week to sort out setup issues. Full setup includes a month of support for the jobs we built. After that, you can run it from the handover notes or arrange ongoing help separately.
Tell me what you're trying to get done and where the information lives. We can work out whether an autonomous agent would earn its keep.
