Skip to content

An assistant that can actually do the work

Not a chatbot bolted onto your website. An assistant built into your own system that can look up any record, draft the email, build the summary and prepare the next action, with a permission system that lets it read freely but write only with your approval. It works your data, on your terms.

Illustration of an AI assistant chat that looks up records and waits for approval before acting

The problem

AI that cannot touch your systems is a toy

Copying data into a chatbot and copying answers back out is not automation, it is a new kind of typing. For AI to earn a seat, it has to operate the actual system, safely, with the same permissions and audit trail as a person.

That is what we build. The assistant sits inside your system with a toolset of real actions, look up a client, summarise a job, draft the reply, prepare the invoice. Reads run freely. Writes pause for a human yes. Destructive actions ask every single time. Every action it takes is logged like any staff member's.

In production today, an assistant like this runs scheduled routines each morning, catches things that need attention, drafts responses for review, and hands work back as tasks. Its token spend is metered per user, so the bill is never a surprise.

This is a module of our custom ERP platform, not a standalone product. Start with just this module and it already sits on the full platform, accounts, security, admin and API included, so the rest of the system can arrive stage by stage, whenever you are ready.

What you get

A colleague, with a permission slip

It operates the real system

Dozens of gated actions across your records, not a script with a personality. Ask in plain language, watch it look things up and prepare the work.

Reads free, writes approved

Looking things up happens instantly. Anything that changes data pauses for your approval, and the risky category asks every time, no exceptions.

Scheduled routines

Every weekday at 7am: check overnight enquiries, flag stalled jobs, draft the follow-ups. Results arrive by email, with anything needing judgement handed back as a task.

Meet it where you are

Docked beside every screen, aware of the record you are viewing, and reachable by replying to its emails when you are away from the desk.

Metered, not mysterious

Every AI call is tracked with its cost, per user and per conversation, with monthly limits you set. The bill is a dial, not a shock.

Guarded against manipulation

Content from forms and emails is treated as data, never as instructions, so a cleverly-worded enquiry cannot trick the assistant into acting on it.

AI native

Run it from Claude or ChatGPT

Every module ships on a platform with a built-in MCP server, which means AI assistants like Claude and ChatGPT can securely read and operate your system with the same permissions as a staff member. Ask where a job is up to, get a client summary before a call, or have tomorrow's follow-ups drafted, from the chat window you already use.

FAQs

AI assistant questions, answered

We build on Anthropic's Claude models by default, with API terms that exclude training on your data. The assistant sees only what the current user is allowed to see, requests are not used to train models, and the system prompt, permissions and logs all live in software you own.

Give your system a colleague

Book a free 15-minute call. Tell us the questions your team answers all day, and we'll tell you what a build costs.

Australian-hostedPrivacy Act compliantNDAs standard