One platform for your knowledge, your AI co-workers, your workflows and your tickets. Your team builds what it needs without code. AI does the repeatable work, and a person approves what matters.
Eva drafted a reply to ticket 1042Waiting for your approval
Approve
Built in Sydney
People approve. Everything is recorded.On your premises or your cloud
Built for organisations including
NSW Health
NSW Department of Education
NSW Telco Authority
Optus
What is an AIOS
Most companies have AI tools. Few have an AI operating system.
An AI operating system is the layer that connects what your business knows, the AI models you use and the work you do every day.
A computer's operating system manages memory, files and devices so that every application can share them. An AIOS does the same job for AI.
It holds your knowledge, connects to the systems you run, and gives every AI co-worker and workflow the same context, the same connections and the same rules. The AI works from what your business already knows, and it can act on it.
Separate AI tools
Each tool knows nothing about the others
Every conversation starts from zero
People paste the context in by hand
It can answer, but it cannot act
No single record of what the AI did
Radmis AIOS
One knowledge base, shared by every co-worker
Answers come from your documents, with the source
Connected to your tickets, email and data
Workflows do the work and wait for approval
One audit trail for people and AI alike
How it is built
Six layers. One platform.
Every AI operating system needs these six. In Radmis they ship together, already connected.
Work
Where things get done: requests, screens and repeatable steps.
In RadmisTicketing, Builder, Co-workers and Workflows
Knowledge
What your business knows: policies, manuals, contracts, history.
In RadmisRAG stores, selected into the Knowledge each co-worker answers from
Models
The reasoning. Matched to the job, and replaceable.
In RadmisLocal or frontier models, chosen per co-worker
Connections
The way in and out, so AI can act and not only answer.
In RadmisMCP and API connectors, and email
Control
Who may do what, who approved it, and the record of it.
In RadmisAccess levels, approvals, audit trail and a kill switch
Platform Health
The platform checks itself in real time, so a fault is found before your people find it.
In RadmisReal-time and scheduled self-checks on one screen, with history and email alerts
Why it pays
What your business gets. Beyond a chatbot.
01RAG and Knowledge
Answers from your own documents.
Upload policies, manuals and contracts. Co-workers answer from them and point to the page the answer came from.
02Builder
Screens built in minutes.
Describe a register, a form or a report in plain words. The platform builds it with access control and audit included. No code is written.
03Ticketing and Workflow
Less manual handling.
Incoming email becomes a ticket and can be filed automatically. Steps that repeat run as workflows.
04Co-workers
An AI colleague for every role.
Set up a co-worker for each job: answering questions from your documents, drafting replies, describing new screens. Each has a name, an owner and an access level, so you always know who did what.
05Models
Your choice of model.
Each co-worker uses the model you pick: a local model on your own hardware, or a frontier model. Changing it is one setting.
06Governance
Control without friction.
People approve what matters, each AI has an access level, and everything is recorded. Governance comes with the platform, so the work keeps moving.
@record
@table gifts
'Received by' fullname*: text
'From' giver*: text | 'Value' amount*: money
'Decision' decision [select: pending, kept, returned]
Gifts register
Builder The platform that builds itself.
Screens without code.
Describe a list, a form or a report. The platform builds it from a short text file called a Sketch. An AI co-worker can write that file for you, you can describe what you need in an email, you can write it by hand, or the Builder configurator can make it for you. All of that without AI ever being in your production platform or having access to your data.
Four weeks for more than five days. Shorter leave needs your manager's agreement. Leave policy.pdf · page 12
Knowledge
Your documents, ready to answer.
Upload PDFs, Word files and more into RAG stores, where they are indexed automatically. A Knowledge is a selection of RAG stores. Give one to a co-worker and it answers from those documents, with the source page.
RAG stores
PDF, Word and other formats, indexed on upload. No preparation needed.
Sources
Each answer names the document and page.
Your storage
Each RAG store is kept in Chroma, Postgres or SQL Server.
Knowledge
A selection of RAG stores. You choose which Knowledge each co-worker uses.
A co-worker is a named AI with its own role, model, knowledge, tools and guardrails. It has an account and an access level, like anyone else on the team.
One email, handled end to end. With a person in charge.
An email arrives
It becomes a ticket in the queue.
It is filed
The classifier picks the category and shows how sure it is.
A reply is drafted
A co-worker writes it from your knowledge, with sources.
A person approves
Email written by AI is not sent without it.
Sent and recorded
The reply is marked as AI-drafted. Every step is in the trail.
One example. Your workflows use the same parts in the order your work needs.
Regulated environments
Usable by your team. Defensible to your auditor.
Radmis comes from health, government and telecommunications, where every system has to pass an audit. The same controls are part of the AIOS. They answer the five questions a risk review asks.
01Who owns it?
The AI Register holds an owner, a purpose and a risk rating for each AI system. It is approved before it goes live.
02What may it do?
A co-worker is a user with an access level. It reaches your data only through calls you have registered.
03What did it do?
Changes to audited data are recorded in a chain of SHA-256 hashes. Each AI turn is kept with its sources.
04Who checked?
A workflow waits at its approval step until a person decides. The decision is recorded with their name.
05How do we stop it?
One kill switch closes the connectors and the models. Each use is logged and emailed.
Your premises or your cloud
Install it where your data already lives. Local models keep sensitive work inside your network.
Personal data protected
The sanitiser can mask names and personal details before text reaches an outside model, then restore them in the answer.
Mapped, not certified
Thirteen regulations and guidelines mapped point by point on 28 October 2026, including APRA CPS 230 and CPS 234, the Privacy Act 1988, NIST AI RMF, the EU AI Act and GDPR.
Six people have a say in AI. It has to work for each of them.
Example personas from the kinds of organisation we build for. The names are illustrative.
Uses it every day
D
DanOperations lead
The job
Runs compliance operations at a finance company: complaints, registers and case work.
The problem
Thursday afternoon, the auditor asks for the gifts and hospitality register by Monday. It is a spreadsheet last updated in March, three people keep their own copies, and IT's queue for a proper system is six months long.
With Radmis AIOS
From the train, Dan emails Eva, an AI co-worker: the fields, who decides, who sees what. Four minutes later the register has a form, a list the auditor can export and a report for the audit committee, with access levels, personal data marking and the audit trail already on.
Describe it in an email. Use it by afternoon.
Runs technology
M
MarcusCIO, Head of Digital
The job
Deliver more with the same team and a two-year backlog.
The problem
Three AI pilots worked in the demo: a policy chatbot, a claims summariser and an inbox sorter. Three vendors built them on three stacks, all three are stuck at the risk committee, and none has gone live. Meanwhile the queue of small internal apps keeps growing.
With Radmis AIOS
All three run on one platform. The chatbot answers from approved documents and cites them, the sorter runs on a local classifier, and the summariser is a workflow a person approves. Each reaches the risk committee with a register entry, test results and an off switch. The small apps get built from a plain-English request on the same platform, which installs in about an hour.
Governed AI and internal apps in minutes, not quarters.
Signs off the risk
R
RachelChief Risk and Compliance Officer
The job
Keep the organisation out of the regulator's headlines and the board out of trouble.
The problem
Customer service wants an AI to draft replies to complaints, and the board meets in three weeks. Rachel is asked to sign it off, but nobody can tell her who owns it, what it may say, who checks the drafts or how to stop it. If it goes wrong, the regulator calls her.
With Radmis AIOS
The use goes on the AI Register with an owner and a risk rating, and stays behind the go-live gate until she approves it. No AI-written reply goes out until a person approves it, and each one is marked as written by AI. A hash-chained audit trail records what happened, and her auditor can verify it without us. One kill switch stops it.
Say yes to AI, with the evidence to prove it.
Guards security
P
PriyaChief Information Security Officer
The job
No breach, no data leakage, no agent doing something it should not.
The problem
Case officers paste client files into a public chatbot because the approved tools are slow. One vendor wants to give an agent direct access to the database, and another wants AI to write the integration code. Either way, customer records reach a model she does not control.
With Radmis AIOS
Staff get a chat that answers from approved documents, so the pasting stops. AI comes in only through the MCP and API gates, as a user with an access level, and every call is checked and logged. It never touches the database directly and never writes code. Before text goes to an outside model, the sanitiser can mask personal details. One kill switch stops the gates and the models.
AI stays outside. Your data stays in.
Accountable in government
H
HelenAccountable AI Official, public sector
The job
Meet the government AI policies: register AI use, assess its impact and publish transparency statements.
The problem
The agency's transparency statement is due, and it must say where AI is used. AI has turned up in a dozen places: translation in the call centre, a summariser in case management, a chatbot on the website. There is no list, and the impact assessments are Word documents in someone's inbox.
With Radmis AIOS
Every use goes on the AI Register with an owner and a risk rating, outside tools included, and the Register flags any co-worker nobody registered. Impact assessments are kept on tickets as evidence, reviews come round on schedule, and the transparency reports are produced from the same records. It runs on the agency's own premises.
Turn AI policy into practice, and prove it.
Owns the AI estate
B
BobHead of Technology
The job
Keep the company's systems running, and make the AI already in them pay off.
The problem
AI came in from every side: Copilot in Microsoft 365, the AI inside SAP, agents an integrator built, and teams on their own subscriptions. Nobody can say what runs, who owns it or what data it sees.
With Radmis AIOS
One register over the AI he already has, each use with an owner, a risk rating and a review date, and every supplier in the providers register with an exit plan. Every model sits behind a label. Outside AI tools reach the platform only through the two gates, and every call is logged.
Every AI you run, on one register, with an owner.
Worked examples
What they do with it. Six jobs, start to finish.
Each example names the part of the platform it runs on. The names are illustrative.
Builder and mail workflow
G
GraceAccounts payable lead
The job
Pay suppliers on time, and only for what was ordered.
The problem
About 600 invoices a month reach the accounts inbox, and a third are missing a PO number, an ABN or the GST line. Two clerks spend their mornings emailing suppliers and chasing answers, and nobody can see which invoice is waiting on whom.
With Radmis AIOS
Grace describes an Invoices register in Builder Chat. Every mail to the inbox runs a workflow: Laya tells invoices from statements and reminders, and a co-worker checks what is missing, asks the supplier in the same email thread and waits, asking again until the invoice is complete. Then it goes before Grace to approve. Every question and answer stays on the ticket.
Incomplete invoices chase themselves.
Sanitiser
M
Dr MeiClinical director, medical practice group
The job
Better care from the same appointment time.
The problem
GPs want a frontier model to summarise two years of a patient's pathology results and draft a letter in plain English. The results carry the patient's name, address and Medicare number, and health information cannot go to an offshore model as it is.
With Radmis AIOS
The Results chat runs with a sanitiser. Before anything leaves, the name, address, Medicare number, phone and email become labels such as [PERSON_1] and [MEDICARE_1]. Names and addresses are found by a model on the practice's own server, and if it cannot be reached, nothing is sent. The frontier model sees the results and the labels. The letter comes back with the real details restored, and the mapping is kept encrypted for that conversation only.
The best model on your results, without your patients' names.
Knowledge with citations
S
SamContact centre manager, insurer
The job
Right answers on the first call, from a team that changes every quarter.
The problem
Forty product disclosure statements of over 100 pages each. New staff guess or ask the person next to them. Last quarter two customers were told the wrong waiting period, and one complaint reached the ombudsman.
With Radmis AIOS
The statements go into a RAG store by drag and drop, or in bulk from a watched folder. The Policy chat answers only from them and cites the document and page. A guardrail checks each answer and refuses anything that reads as personal advice. Before go-live, a test set runs 50 known questions. Staff can report a wrong answer, which opens an AI incident. A new version of a statement goes into the store, and every chat answers from it.
Every answer from your documents, with the page it came from.
Personal data requests
L
LenaPrivacy officer
The job
Answer every privacy request in time, and be able to show how.
The problem
A former customer asks for everything held about them, then asks for it to be erased. That means two weeks of searching the CRM, shared inboxes, chat logs and file shares while the 30-day clock runs, and some of it is under a legal hold.
With Radmis AIOS
One search covers tables, conversations, mail, files and logs. From the results she exports, erases or restricts, and a record under a hold is left untouched. Each step is recorded on the request, and Do Not Mail keeps the person off future mailings.
Thirty days becomes an afternoon, with the record to prove it.
Triage on your own server
T
TomComplaints manager, energy retailer
The job
Every complaint to the right team, and the vulnerable customers first.
The problem
After a price rise, 3,000 emails arrive in a week. Hardship cases sit in the same queue as billing questions, and missing one is a breach.
With Radmis AIOS
Every email becomes a ticket. Laya, the classifier on the company's own server, sorts each one by type, scores its urgency and answers one question: is this customer in hardship? A yes files the ticket under the hardship subtype, and its workflow sends it to the specialist team at once. No email leaves the building to be sorted.
Thousands sorted in seconds, the urgent ones first.
Audit trail
K
KateHead of Internal Audit, credit union
The job
Give the board and the external auditor evidence they can rely on.
The problem
The external auditor asks her to prove that nobody changed the hardship register after a decision was made. The administrators can edit any row, and the only log sits in the same database they can change.
With Radmis AIOS
One tick on Audit Setup, and the database itself records every change to the register, whoever makes it: who, when, the old value and the new, in a SHA-256 hash chain. Verify recomputes the chain, so an entry altered afterwards shows. Her auditor runs the check without us. Settings Changes keeps every settings save too, with who, when and both values.
Evidence that shows if anyone rewrites it.
Track record
Twenty-six years of systems that had to pass an audit.
Every capability in the platform was built for a real system first.
26 years
of research and development behind the platform
100+
deployments of data management systems and services
3,000+
telecommunications sites audited on one platform
200+ hours
of manual data entry saved in radiology
Telecommunications · infrastructure audit
3,000+ sites. Three auditors instead of eight.
Problem
No system for the job, and several vendor products that had never been integrated.
Built
An interim audit platform with secure mobile tools and automated compliance reporting.
Result
The interim system did the work, and became the working model for the permanent one.
As reported by the Network Infrastructure Program Lead, telco operator
Government · Department of Education
Weeks of reconciliation, now running in the background.
Problem
A limitation in the legacy ERP, and several systems that did not talk to each other.
Built
A connection between those systems using AI, automating policy-driven adjustments for thousands of employees.
Result
Reconciliation that took weeks now runs in the background, with fewer errors.
As reported by the Director, Workforce Policy Division
Technology partners
Questions
Asked often. Answered plainly.
What is an AI operating system?
It is the layer that connects your business knowledge, your AI models and your daily work. Separate AI tools each start from nothing. In an AIOS every co-worker and workflow shares the same knowledge, the same connections and the same rules.
How is it different from a chat assistant?
A general chat assistant answers questions. It knows nothing about your business unless someone pastes the context in, and it cannot act in your systems. Radmis AIOS answers from your documents, works your tickets and workflows, and keeps a record of what it did. You can still connect a frontier-model assistant to the platform through MCP.
Do we need developers?
Not for everyday building. Lists, forms and reports are described in plain text, and an AI co-worker can write that description for you. Workflows are readable text files too. Your IT team installs the platform and decides who may do what.
The Builder configures new functions. It does not generate code, so there is no code to review and no rigorous testing to do. If a Sketch file is not correct, the Builder refuses to build and asks for changes. That makes it safe to use on a production server, to make new forms or change existing ones.
Which AI models does it use?
You choose for each co-worker. Today that means local open models running on your own hardware, or a frontier model. Changing a co-worker's model is one setting, so you are not tied to one supplier.
Where does it run, and where does our data go?
On your premises or in your cloud. Your data stays in your own database. With local models, nothing leaves your network. If you use an outside model, the sanitiser can mask personal details before the text is sent and restore them in the answer.
Can we use it in a regulated environment?
That is where it comes from. Every AI has an owner, an access level and an audit trail. People approve what matters, and one switch stops it. The controls are mapped point by point to thirteen regulations and guidelines. They are mapped, not certified. See the governance page
How do we start?
With one piece of work. Installing takes about an hour, and a first register can be on screen the same day. Book a demo and we will build a working screen live while you watch.
Bring one piece of work. See it run.
We build a working screen live in the meeting, then look at where an AI operating system would help your team first.