Custom AI agents for business teams

A custom AI agent is a teammate your staff can message in Google Chat or Slack. It knows your business, reads your own systems and does real work in them: looking up an order, checking stock, drafting a reply or changing your website. Anything that matters waits for a person to say yes. We build them for businesses in Sydney, across Australia and overseas.

Case studies

Related work

Shroomita

An AI agent the whole Grolia team messages in Google Chat. It checks customer cabinets, turns faults into fixes, makes ad images and builds features for review.

Wahlu

Social media scheduling software with a content calendar, evergreen queues, AI planning and a hosted MCP server for AI agents.

What we build

Custom AI agents

  • An agent in your team chat

    People tag it in a Google Chat or Slack thread, ask in plain English and get the answer in the same thread. Follow-ups keep the context.

  • Connected to your systems

    Orders, customers, stock, device data, email and shared drives, with access you control system by system.

  • Changes with approval

    You decide what it may do on its own, what needs an explicit yes and what it must never touch. Your system enforces those rules, not just the prompt.

  • Website and code changes

    For teams with software, it can make a change, open a pull request and share a preview for someone to approve. Merging and deploying stay with your developers.

  • Images and documents

    Product images, draft copy and summaries, saved to a shared folder for the team to use or reject.

  • An audit trail

    Every change is recorded against the person who asked for it, so you can always see who did what through the agent.

Approach

How we build it

We start with the questions your team asks one person all day, and the jobs that wait for that person to be free. Then we agree, system by system, what the agent can read, what it can do, what needs a yes first and what it must never do.

At Grolia, Steve built Shroomita, an agent the whole team uses in Google Chat. It runs on Claude through Anthropic’s Agent SDK. Its admin changes go through Grolia’s own admin API, which checks every request’s permission level and logs who asked.

Questions

Common questions

What is a custom AI agent?

An AI model connected to your business’s own systems and rules, so it can act as well as answer. A general assistant can explain how to check an order. A custom agent can look the order up, see that it is stuck and fix it, if you have allowed that.

How is it different from ChatGPT Business or Gemini in Google Workspace?

Those are good at writing, summaries and searching your documents. A custom agent also works in the systems specific to your business, such as your order admin, stock, device data or website code, and follows approval rules you set for each kind of change.

Is it safe to let an AI agent change things?

It is safe when the limits live in your systems. Each action is allowed, allowed after an explicit yes, or blocked, and the server checks this on every request. Customer emails, payments and deployments can stay off limits entirely, and every change is logged.

Do our staff need to be technical to use it?

No. They ask in plain English in the chat app they already use. At Grolia, the design, manufacturing and creative leads all use the agent, and it explains its work in everyday language.

Google Chat, Slack or Microsoft Teams?

Whichever your team already uses. Shroomita runs in Google Chat, and the same design works in Slack or Teams. Only the connection to the chat app changes.

Do you build AI agents for small businesses?

Yes. Grolia is a small team, and that is where an agent helps most: there is no IT department, so one person usually ends up answering everyone’s system questions.

How much does a project cost?

Projects start from US$12,000, quoted in US dollars. The enquiry form has a budget field, so tell us the range you have in mind and we will say whether it fits.

Working together

How an enquiry works

No sales calls to book. Tell us about the project and we’ll take it from there.

  1. Send an enquiry

    Tell us what you need built, what you use now and roughly when. A few sentences is enough to start.

  2. Steve reads it

    Every enquiry is read by Steve, the founder. If it looks like a good fit, he replies by email with questions.

  3. Talk it through

    If the project suits us both, we talk through what to build first and how to get there.

Want an AI agent for your team?

A few sentences is enough. Steve reads every enquiry and replies by email.

Send an enquiry