The questions people actually ask

It logs the failure, stops, and flags it as "Failed" with the reason. Nothing retries silently and nothing half-completes. If something did go out wrong, the activity log shows exactly what the agent read and decided, so you can correct it in minutes rather than reconstructing it.

Only if you've set it to. Every agent starts in shadow mode, where it sends nothing at all. You move it up a level at a time, and the limits you set — value, new customers, anything tone-sensitive — always hold.

That's your decision, not ours. What we see most often is the same people doing the work they were hired for instead of the paperwork that accumulated around it. We'd rather tell you that than promise headcount savings we can't stand behind.

It's credited in full against your first month if you go ahead. If you don't, you keep the audit — the findings are yours either way, and plenty of people have used them to fix things themselves.

Credits meter the work your agents do — an invoice drafted, a lead qualified, a report written. Your plan includes a monthly allowance and the dashboard shows what's left. We tell you before you hit it, not after.

No. Model and API costs sit inside your credit allowance. You get one predictable monthly number, which is the whole point of buying it as a service.

Every plan runs for a minimum of six months after your setup month. That commitment is what lets us put real work into building your agents properly rather than rushing it. After six months it rolls monthly with 30 days' notice, and annual is 20% cheaper.

You own your data and everything the agents produced, and you can export both. The agent logic and prompts stay ours — that's what the subscription buys access to. If you need to own the system outright, that's a custom build instead.

No. Your tenant is isolated, and nothing you put in improves a model that serves another customer. Security & data has the detail your IT person will want.

Month 0 is setup — shadowing your process, connecting systems, running in shadow mode. Agents go live at the end of it, at low autonomy. Anyone promising you live agents in week one hasn't looked at your data yet.

Because a framework is not a product. The open-source tools are free to download and genuinely good, but they ship without a brain — you supply the AI itself, which means your own developer account, your own API key and your own credit balance. Then you still need somewhere to run it, a dashboard to see what it did, a user interface your staff will actually use, the integrations into Xero and WhatsApp, and someone maintaining the server at 11pm on a Sunday. Agentic Labs is all of that, already built. You subscribe, and our team builds the agents on top of it.

No. The credits are ours and they are in the price. You will never be asked to create an account with a model provider, paste a key into anything, or watch a usage meter. If you would rather hold your own provider account, we can work that way instead — but nobody has asked yet.

Yes. That is the custom build: your own system, on your own infrastructure, with the code and the logic yours forever. It is a lump sum rather than a monthly fee, and it makes sense when your processes are unusual enough that a subscription would be fighting them. Most businesses start on the subscription and a few move to a build later.

We'll tell you on the discovery call rather than three weeks in. Legacy and on-premise systems get scoped separately.

We are Singapore-based and every client so far is here, which is deliberate — knowing how an IRAS deadline or a progress claim actually works locally is half the value. We will take a conversation from elsewhere in the region, but we will say plainly where our knowledge of your local rules runs out.