Most software makes your AI agent scrape screens and guess. Mazgan hands it real tools — jobs, customers, equipment, estimates, invoices — with one rule above all: nothing reaches a customer without your approval.
New lead from your website: boiler, no heat, Vaughan. I've booked the visit for tomorrow 9 AM and drafted the confirmation. Awaiting your approval.
Approved. Send it.
Estimate prepped from the job record, equipment history attached — the boiler's last combustion readings are on file. Drafted — awaiting your approval.
An agent is only as good as the tools it's given. These are the tools — designed for an agent first, screens second.
Typed actions for jobs, customers, equipment, estimates, invoices and payments. No screen-scraping, no brittle clicking — your agent works with real operations.
The moment your agent connects, it reads one brief: your business, your rules, your document standards. Oriented in minutes — not weeks of prompt archaeology.
Every write can be rehearsed first. Idempotency and expected-state checks mean your agent can't double-book, double-charge, or quietly overwrite your records.
Your agent drafts the message, the estimate, the schedule change. You tap approve — or edit — before anything reaches a customer. Ever.
Invoices are born linked to their jobs. Documents carry your words byte-for-byte. Records archive instead of vanishing. The log says what actually happened.
A per-business credential, revocable any time. Your agent sees your business — never anyone else's — and every action it takes is written to an audit trail.
Mazgan doesn't lock you to one AI. Connect the agent of your choosing with your business credential — it works under your rules, in your CRM.
Speed is worthless if it sends the wrong thing to a customer. So the system is built around a single gate: agent work is a draft until the owner says go.
Messages, estimates and changes wait for your tap. In the first release, the agent doesn't send customer messages at all — it prepares them for you.
Prices, readings and job details come from the record or stay marked Pending. Your agent never fills a gap with a guess.
Who did what, when, and what it changed — an audit trail you can actually read, for agent actions and manual ones alike.
“Hi Maria, confirming tomorrow 9 AM for your boiler service. Reply here if anything changes.”
Built from the job record, your wording carried over exactly. Total $1,240 + HST.
No conflicts found. Customer not yet notified — draft message attached.
A lead becomes a customer and a job in the CRM. From there, every agent change is previewed before it is saved — and customer messages stay drafts in your hands, never auto-sent.
Jobs booked, records kept, paperwork drafted, follow-ups queued. You see proposals, not chores — approve from the truck, the roof, wherever you are.
The back office stops eating your evenings. Every record is straight, every customer got an answer, and you approved all of it.
The public version — Foreman — is in final build. Join the list and you'll hear first when doors open, and what your agent will be able to do on day one.
The agent of your choosing. Mazgan is built API-first with a written agent contract, so Muse, ChatGPT-based agents, Codex, or your own build can connect with a per-business credential and work under the same rules and protections.
Not on its own. In this release the agent cannot send customer messages at all — it drafts, and you send from your own apps. Document changes go through a sealed preview you confirm first, and payment links follow a policy you control.
Your business is its own tenant: its own records, its own credential, its own audit trail. Records are archived, never silently deleted, and documents keep your exact wording.
The public version, Foreman, is in final build and testing. Join the interest list above and you'll be first to know when early access opens — and your place in line is recorded in the very system you'd be using.