AI in ComplyFlow

AI That Reads Against Your Criteria, and Shows Its Working.

A 40-page safety manual graded against your requirements in seconds, with the reasons on the record. Agents you build yourself. An assistant that answers from the live data and cannot change it.

Book a Demo See the Three Jobs

Safety & HSE
Procurement
Contractor Admins
Operations
IT & Data

AI Review (2 remaining) Download feedback

B Meets requirements

Meets the minimum standard you set for this document type.
Minimum required: B. Emergency Response needs work before the permit is approved.

Section grades

Section Grade Score

1. Scope of Works A 92%

2. Hazard Identification B 78%

3. Control Measures B 75%

4. Emergency Response C 58%

5. Sign-off & Review B 81%

AI usage included in Enterprise plans since 24 March 2026

What AI in ComplyFlow Does.

AI in ComplyFlow does three jobs. It reviews the documents contractors submit against the criteria you set and records the grade and reasons. It drafts method statements, checklists, and inductions from your material. And through MCP it answers questions from your live record without changing it.

Review. A safety plan, SWMS, certificate, or prequalification answer read against the criteria you set, and graded with the reasons.
Draft. A method statement from a template for the work, an inspection checklist that fits the site, an induction from your own material.
Ask. Claude, Claude Code, or GitHub Copilot reads your incidents, inspections, suppliers, Workers, risks, plant, sites, and permits through MCP. Read-only.
You build the agent, without code. Pick a form type, choose the questions, keep or edit the criteria the product drafts for you.

A safety manager at a site office desk reading a thick printed safety manual beside a laptop, with a pen in hand and sticky notes on the pages

How the AI Works

Your Criteria in the Middle. Documents In. Decisions Out.

Read it left to right: what your people and contractors put in, the criteria you set that everything is read against, and what comes out for a person to act on.

What goes in

A safety plan or method statementUploaded by the contractor against a requirement review

A prequalification answer and its evidenceA text, text area, or file upload question with an agent on it review

A licence, ticket, or insurance certificateThe document a Worker or supplier submits

Your own procedures and documentsThe material a course or induction is drafted from draft

A question in plain EnglishAsked in Claude, Claude Code, or GitHub Copilot ask

What comes out

A grade per requirement, with the reasonsOn the document, for the approver and the audit

A score on the submissionThe reviewer sees the answer and the assessment together

The expiry read and the approver’s queueThe approver still decides; the AI recommends

A draft SWMS, checklist, or courseFrom a template for the work, edited by you draft

The answer, from the live recordRead-only, with the user’s own permissions ask

The AI reads against criteria you set, per form type and per requirement. Nothing it produces changes a record on its own; a person approves, rejects, publishes, or closes.

What It Covers

Three Jobs: Review, Draft, Ask.

Everything below is in the product today. Nothing here is a roadmap.

Review

AI document review

A document graded against your requirements, with reasons, before approval.
- Per requirement
- Graded
AI agents on form questions, opens the Help Centre in a new tab

An agent reviews a text or file answer as it is submitted.
- Per form type
- Several per question
AI prequalification

A supplier’s questionnaire and evidence assessed against your criteria.
- Suppliers
Criteria drafted for you

The wizard suggests criteria; you keep, edit, or add your own.
- Wizard
- Custom criterion
The decision on the record

Grade, reasons, and criteria stay with the document for the audit.
- Audit trail

Draft

SWMS review and templates

A method statement drafted for the work, or an existing one reviewed.
- Templates
- Review
Inspection checklists

A checklist that fits the site, with corrective actions suggested.
- Per site
- Actions
Course and induction creator

An induction drafted from your procedures, then published in Training.
- Training module
Custom agents

An agent for any review you run: form, questions, criteria.
- No code
Feedback the contractor can act on

The gaps and what would meet each criterion, downloadable.
- Download

Ask

Your assistant, connected, opens the Help Centre in a new tab

Claude, Claude Code, or GitHub Copilot through the MCP server.
- MCP
- Read-only
Eight modules to ask about, opens the Help Centre in a new tab

Incidents, inspections, suppliers, Workers, risks, plant, sites, permits.
- Live data
Scoped, expiring tokens, opens the Help Centre in a new tab

Read scopes only, expiring within a year, revoked on confirmation.
- cf_pat_
- Up to 10
Your permissions, not more, opens the Help Centre in a new tab

The assistant sees what your user and assigned sites see.
- Inherited
Included in Enterprise plans, opens the Help Centre in a new tab

AI usage included since 24 March 2026; others pay as they go.
- Since 24 March 2026

Where the Help Centre documents an item, its name links to the article.

How It Works

Four Screens That Carry the Feature.

An agent built, a form answer reviewed, a course drafted, and a question answered from the record.

The Agent You Build in Four Steps.

Name it, choose the form type, choose the questions, pick the criteria. The product drafts criteria for the kind of document it will read; you keep, edit, or add your own.

Criteria suggested by the product, marked as such
A custom criterion in your own words
One agent per form type; several on one question

Select criteria to review against

Step 3 of 4

Selected: 2 / 10Select All

Scope of Works Defined AI The tasks, location, and dates the SWMS covers are stated.
High-Risk Work Identified AI Each high-risk construction work activity is named.
Controls Follow the Hierarchy AI Controls eliminate first, then substitute, isolate, engineer, administer, and PPE last.
Competency Stated AI The licences, tickets, and training each task needs are listed.
Emergency Response AI What happens if a control fails, and who is called.
Sign-off & Review AI Workers have signed on, and a review trigger is set.

+ Add Custom Criterion

The Form Answer, Reviewed as It Lands.

A supplier answers a prequalification question and attaches evidence. The agent on that question scores the answer against the criteria and leaves its assessment on the submission for the reviewer.

Text, text area, and file upload questions
Assessment recorded on the submission
Reviewer sees the answer and the reasons together

Prequalification, Tallis Civil Pty Ltd

Form 2291 · submitted 04 Mar 2026 by Dale Whitford · Under review

Under review

Section 3 · Question 3.4

When was your safety management system last audited, and by whom? Attach the audit report.

Contractor’s answer Internal audit, February 2025, by our WHS coordinator. Report attached. tallis-sms-internal-audit-feb-2025.pdf

Your feedback Action required Please attach the external audit or certification, not the internal one. Section 3.4 asks for an independent audit within 24 months.

The Induction, Drafted From Your Procedures.

Point the course creator at your site rules and procedures. It drafts the sections and the questions; you edit, then publish it in Training and assign it by Worker category.

Drafted from your own documents
Edited and published in Training
Assigned by category; completions chased

Create Course (AI)

Step 2 of 3 · review the outline, then create the course.

Create course

Outline

1 Why isolation matters on this site
2 The five steps of lockout
3 Personal locks and tags
4 Verifying zero energy
5 Handing back and removing locks
6 Quiz: six questions

+ Add a screen

The Question, Answered From the Live Record.

A safety manager asks which sites reported incidents this month. Through MCP the assistant reads the register with the scopes on their token and their own site permissions, and answers with the rows.

Claude, Claude Code, and GitHub Copilot
Read-only, with the user’s own permissions
Tokens scoped per module, expiring within a year

You

Which sites reported incidents this month, and what was the most serious at each?

Assistant

Reading ComplyFlowlist_incidentsget_site

Three sites reported incidents in September. From your ComplyFlow record:

Site	Incidents	Most serious
Marlow Point Terminal	3	Medical treatment
Greyhaven Wharf 9	2	First aid
Wattle Creek Line	1	Near miss

I can read your incidents, inspections, suppliers, workers, risks, plant, sites, and permits with the scopes on your token. I cannot create, change, or delete anything.

Model Context Protocol

The Third Job: Ask. Your Assistant Reads ComplyFlow.

What it is: MCP is the open standard that lets an AI assistant read another system safely. ComplyFlow ships an MCP server, so Claude, Claude Code, or GitHub Copilot can read your compliance record.
What it lets you do: Ask which sites reported incidents this month, who on site is not cleared, or what is overdue on the risk register, and get the answer from the live record, in the assistant you already have open.

Read-only by design. The assistant sees what you may see, at your assigned sites, and cannot create, change, or delete a thing.

The first contractor compliance platform we can find with its own MCP server Read the MCP Guide

You

Which sites reported incidents this month, and what was the most serious at each?

Assistant

Reading ComplyFlowlist_incidentsget_site

Three sites reported incidents in September. From your ComplyFlow record:

Site	Incidents	Most serious
Marlow Point Terminal	3	Medical treatment
Greyhaven Wharf 9	2	First aid
Wattle Creek Line	1	Near miss

I can read your incidents, inspections, suppliers, workers, risks, plant, sites, and permits with the scopes on your token. I cannot create, change, or delete anything.

MCP

Which sites reported incidents this month?
Who is on site today with an expired licence?
Which inspections are overdue, and whose?
Summarise the open risks at Marlow Point

Who It Is For

What AI in ComplyFlow Means for Me.

Find your job. The three lines under it are what the three AI jobs change for you.

ComplyFlow Is Already Trusted.

Documents at property, transport, resources, manufacturing, government sites, and more are already reviewed on ComplyFlow.

Book a Demo

Bring a Document You Dread Reviewing. We Will Grade It in Front of You.

A 30-minute call with someone who has set this up for procurement and safety teams: your requirements, one of your contractors’ documents, the grade and the reasons, live.

One of your own documents graded against your requirements, with the reasons
A review agent built for a form question in four steps
Your assistant asking the live record through MCP, read-only

Talk to Us Instead

No slides, no hard sell. A working session on your own set-up.

Book Your Demo

Thirty minutes, on your own requirements.

Book Demo

The form loads from HubSpot. If it does not appear, the button takes you to the demo request page.

A stack of contractors’ safety documents and method statements in ring binders on an office desk, sticky notes and a highlighter on top, a monitor asleep behind

How It Stacks Up

Where AI Review Usually Lives, and What Each Misses.

Most teams read documents themselves, paste them into a general chatbot, or use AI bolted onto a system that does not hold the record.

	Reading it yourselfThorough on a good day; 40 pages at 4:30 on a Friday is not	A general chatbotKnows nothing about your requirements or your contractor	AI bolted onto another systemThe result lives somewhere other than the record	ComplyFlowOne record, your standard
Reads against your requirements, not a generic standard	Partly	Partly	Partly	Criteria you set, per form type and per requirement
The grade and the reasons on the Worker or supplier record	Partly	Not there	Partly	On the document, visible to the approver and the audit
Consistent from the first document to the thousandth	Not there	Partly	Built in	The same criteria, every time
You build the agent without a developer	Not there	Partly	Partly	A four-step wizard; criteria drafted for you
Answers questions from the live compliance data	Not there	Not there	Partly	MCP into Claude, Claude Code, and Copilot, read-only
Cannot change your data on its own	Partly	Built in	Partly	Read-only tokens; the approver still decides

This compares kinds of approach, by how they are sold, not named vendors. Named comparisons, with dates and sources, live on the comparison pages.

What You Get

What the AI Delivers, in Plain Words.

Fewer hours reading, the same standard applied to every document, and answers without opening a dashboard.

A safety manager in a site office leaning back from two monitors, typing a question on a laptop, with a hard hat and a two-way radio on the desk

The 40 Pages, Read Before You Open Them

A grade per requirement and the passages that earned it. Your approver decides in minutes.

The Same Standard, Every Document

The criteria you set are the criteria applied, on the first submission and the thousandth.

Agents Built by the People Who Know the Work

A safety manager builds a review agent in four steps. No developer, no ticket.

Drafts, Not Blank Pages

A method statement, a checklist, an induction: drafted from your material, edited by you.

Answers in the Assistant You Already Use

Which sites had incidents, who is not cleared, what is overdue: asked in Claude or Copilot.

Nothing Changed Without a Person

The assistant is read-only. Review agents recommend. Your people approve, reject, and close.

Getting Started

We Set It Up With You. We Stay With You.

You are not handed a login and left to it. Our onboarding team turns your requirements into criteria, builds the first agents with you, and stays on hand once the first documents are graded.

1 We Read Your Requirements
What each type of Worker and supplier must hold, and what a good document looks like to you, becomes the criteria the AI reads against.
2 We Build the First Agents
Document review on the requirements that take your team longest, and an agent on the prequalification questions that matter most.
3 We Connect Your Assistant
A scoped token, the MCP server configured in Claude Code or Copilot, and the first questions answered from your data.
4 You Go Live, We Stay Close
Training for your approvers, a Help Centre for everyone, and a support team that picks up the phone.

ISO 27001Certified 2019
Certified 2023
Compliant 2020
Single sign-on

Security & Data

Certified, Audited & Hosted in Australia.

Your compliance record is the evidence you rely on when somebody asks, so where it lives and who can reach it matters. ComplyFlow is ISO 27001 certified, runs on AWS in Australia, and your people sign in with the accounts they already have.

Questions

Questions People Ask About the AI Before a Demo.

Does the AI approve documents on its own?

No. It reads a document against your criteria and records a grade with the reasons. Your approver sees both alongside the document and decides. Agents on form questions assess; a person decides.

What does it read against?

Your criteria. When you build an agent the product suggests criteria for the kind of document it will review, marked as drafted by AI, and you keep, edit, or replace them. The founders describe the review routes as database driven: what the AI checks is what you configured.

Which AI assistants can connect, and what can they do?

Claude, Claude Code, and GitHub Copilot, through the ComplyFlow MCP server, and any other assistant that speaks MCP. They can look up and summarise incidents, inspections, suppliers and contractors, Workers, the risk register, plant, sites, and permits, within the token’s scopes and the user’s permissions. They cannot create, change, or delete anything.

How is the connection secured?

Four layers, as the Help Centre lists them: a personal API token that expires within a year and is revoked the moment you confirm it; scopes that limit the token to named modules, all read-only; the permissions and site assignments of the staff user who created it; and isolation between organisations.

What does the AI cost?

AI usage has been included in Enterprise plans since 24 March 2026. Other plans pay as they go, with usage measured in tokens and shown in real-time in the billing dashboard.

Is our data used to train anything?

That question deserves a written answer from us rather than a line on a web page. Ask for the Information Security Policy and the AI data-handling statement as part of your security review, and we will provide both.

See It on Your Own Documents.

Book a 30-minute demo. Bring a safety plan or a SWMS you would rather not read again; we will build the criteria live, grade it, and show you the reasons. No slides, no hard sell.

Book a Demo Speak to Us

ISO 27001
Hosted on AWS
Microsoft & Okta SSO
API & MCP
Data in Australia

Compliance you can prove, instantly.