AI agents that do the work, not just the talking
An AI agent is software that uses a language model to decide and act, not just chat. It reads your data, calls tools and writes the result back. For a business, that means the repetitive work gets done, not just discussed. An agent can answer a visitor, sort an inbox, update a sheet or prepare an invoice. A person approves what matters. We build agents that work on your own data, with clear limits set from the start.
Who it's for
- Teams answering the same questions on the website, in the inbox or on calls all day
- Back offices copying data between email, Sheets, a CRM and invoices by hand
- Businesses that tried a chatbot and found it could talk but couldn't do anything
What we build
- Voice agents for your website that talk to visitors in the browser
- Website chat agents that look up products and docs and capture leads
- Email and inbox agents that triage, extract details and file them into Google Sheets
- Back-office agents that work on your Sheets, CRM and invoices
- Several agents that hand a conversation or task to each other
- Human handoff, and optional approval steps before actions that cost money
- Guardrails and grounding on your own data, so answers come from your sources
- WhatsApp sales agents (covered on our AI & WhatsApp assistants page)
Proof: agents we run ourselves
Our WhatsApp sales assistant, MageSheet Butik, is an agent that calls real tools. It checks stock, creates and cancels orders, changes a size (and refuses if the order has already shipped), starts returns and sends payment links. When a customer asks for a person, it escalates to the shop manager. It never pretends to be human.
The chat assistant on this site uses tool calls to find products, compare plans, search our docs and save proposal requests.
How we work
- Free discovery call: we pick one task the agent should take over first
- A written scope: what the agent can read, what it can do, and when a person approves
- Build and test on real examples from your business
- Go live with logs, then refine using real conversations and tasks
When this is not the right fit
We'd rather tell you now than sell you the wrong thing.
- You want an agent to make decisions that cost money or are regulated with no human in the loop. We won't build that.
- A simple FAQ page or a few saved replies would do the job. That's cheaper, and we'll suggest it.
- You need it live next week. An agent that takes actions needs testing on real examples before it faces customers.
Frequently Asked Questions
Everything you need to know
A chatbot answers. An agent answers and acts: it uses a language model to decide which tool to use, like looking up stock, updating a spreadsheet or creating an order, then does it and reports back. That's why agents need clear limits and a human handoff, which we design in.
Today we build voice agents that talk to visitors in the browser on your website. Connecting an agent to a phone line is possible, but it needs extra setup and testing, so we scope it on request.
The repetitive, rule-based parts: answering common questions, capturing and qualifying leads, sorting email and pulling out order or invoice details, updating Sheets or a CRM, and preparing documents for someone to approve. Judgement calls stay with your team.
The build is quoted after a short call, because it depends on channels, tools and integrations. For text-based workflows running on Apps Script, the model's API usage is commonly $5–40/month for a small business, occasionally more at high message volume. Voice agents cost more to run; we estimate it in the quote.
We design for it. Answers are grounded in your own data, actions that cost money can require an approval step, the agent hands over to a person when it's unsure or asked to, and every action is logged so you can see what happened and fix the rule.