One assistant on your own computer, connected to your own AI account, set up for one real job, tested with you, and handed over with instructions you can follow.
Approval before anything consequential. The agent asks; a person decides.
They are not the same product, and the difference decides what you are buying.
Answers questions in a browser tab. No access to your files, your email or your systems unless you paste things in. Useful, but it forgets everything when the tab closes.
Runs on your computer or server, uses tools you allow, keeps context between sessions, and asks before doing anything consequential. This is what we install.
Fixed rules that move data between systems without a conversation. Excellent for a defined process, and a different service entirely.
Why the distinction matters: an agent can look like a chat window and still behave like automation, or the reverse. If you are unsure which one you need, the first conversation sorts it out.
One assistant on your own machine, set up properly, doing the small things that eat your evening: research, sorting documents, drafting replies, finding the file you know you saved somewhere.
Device suitability and any third-party costs are assessed before you are quoted. Not every computer is a good candidate, and a model that runs locally is not automatically suitable for the job.
One agent doing defined work with your staff, under supervision: administrative support, document research, internal knowledge, drafting and reporting pulls. Not a replacement for a role, and not something that runs unattended on money or customers.
Standard setup is single-user. Multi-user access, role separation, several integrations or unattended actions are custom work, quoted separately. We do not sell multi-tenant or fully autonomous production environments.
We select the appropriate AI agent and installation method based on your computer, operating system, hardware and intended use. Not every system or AI model is suitable. We confirm compatibility and the proposed setup before work begins.
Hermes Agent on Linux, covering installation, model connection, tool use and handover documentation. This is the configuration we can stand behind today.
Windows installations and container-based deployments. We tell you what applies to your setup, and what we have not yet proven, before you commit to anything.
We currently support Hermes Agent and OpenClaw. Other agent platforms are assessed case by case and may not be suitable, so ask before assuming any particular platform will work. We do not promise that every computer can run a local AI model, and we are not affiliated with, endorsed by or certified by any platform vendor.
One agent, one machine, one tool, one workflow. Written down before we start, tested before handover.
Up to 30 minutes to agree the task, the machine and what "working" means.
Operating system, memory, disk, admin access and intended use, before anything is bought.
One supported agent on one compatible computer, connected to your own provider account or a local model.
One supported tool so the agent can do a job rather than only describe one.
Least privilege, approval before consequential actions, and a spend limit on model usage.
Your own sample tasks, run live, including one request the agent must refuse.
A one-page guide and up to 45 minutes of instruction for the people who will use it.
For defects in the setup we delivered, and it does not limit your Australian Consumer Law rights.
Provider, model, hosting and data stay in your name. Usage is billed by the provider.
Anything consequential asks first. We do not sell fully autonomous production environments.
AI output can be wrong. It is checked before anyone relies on it, and nothing here guarantees a business result.
Not every computer or model is suitable. We confirm the fit before you commit, and changes outside scope are quoted separately.
Routine upkeep for one eligible standard deployment, so a provider change does not turn into a support call you pay for by the hour.
Not included: unlimited or priority support, continuous monitoring, guaranteed uptime, major upgrades, new integrations, third-party usage charges.
Optional. Cancel with 30 days written notice.
Custom deployments are quoted separately for maintenance. Pricing assumes a compatible machine and a supported provider account supplied by you.
If the work involves automating a defined process, connecting several systems, building a skill that does not exist, or running multiple users, it is not a setup job. Those are scoped and quoted as projects.
If you are not sure which applies, the AI Workflow Audit is the cheaper way to find out before committing to a build.
Software that runs on a computer or server, uses tools you approve, keeps context between sessions and carries out tasks under your instructions. It is not a person, it is not always right, and it should not be given control of anything you would not hand to a new staff member on their first day.
A subscription is an account in a browser tab. What we install runs on your own machine, remembers your context, can use tools such as files or documents, and is configured with permissions and approval steps. That configuration is the service.
Sometimes. It depends on the operating system, memory, free disk, administrator access and the model you want to use. We assess that before quoting, and we will tell you if the machine is unsuitable rather than install something that disappoints you.
You need access to a model, either through a provider account you own or a model running locally. Provider usage is billed by the provider, not by us, and it is separate from the setup work. We do not bundle or resell it.
Yes, if the model is suitable. Running in Ollama is not enough on its own: the model has to serve at least 64,000 tokens of context and support tool calling, which rules out most small models by default. Local models also need more hardware than cloud ones.
It can be configured safely, which is a large part of what we do: least privilege, approval before anything consequential, nothing exposed to the internet, credentials stored properly. No setup is guaranteed secure forever, and an agent can still be wrong or be talked into something foolish by a malicious document.
One standard interface or integration, yes. Several systems, multiple users, custom skills or unattended actions are project work and are quoted separately.
Updates, credential rotation and occasional troubleshooting when a provider changes something. Support for defects in the setup we delivered is included for seven calendar days after handover, and that period does not limit your rights under the Australian Consumer Law. After that, AI Agent Care covers routine upkeep, or one-off work is quoted as needed. Ongoing maintenance is optional, not bundled.
Send through your operating system, what the machine is used for and the outcome you have in mind. We will tell you whether a standard setup fits, what it would cost, and what it will not do.
