RadmisAIOS Book a demo

The platform

Six components. One place to work.

Each part is useful on its own. Together they share one knowledge base, one set of connections, one access tree and one audit trail.

The Radmis platform home screen: a menu of home, dashboard, ticketing, co-workers and admin, with cards for ticketing, the site model, users and access, and the audit log.
The platform's home screen

Builder

Describe the screen. The platform builds it.

A register, an entry form or a report is described in a short text file called a Sketch. There are four ways to get one: ask an AI co-worker to write it from a few sentences, write an email that describes the function you need, write a Sketch by hand, or use the Builder configurator to make it for you. The platform reads the file and builds the table and the screens. No code is generated, and AI is never in your production platform or given access to your data.

  • Grids. Searchable lists you can edit in place and export to CSV.
  • Records. Entry forms over any table, with required fields and choices.
  • Reports. Tiles, charts and grouped tables.
  • Menu and access. A new screen takes its place in the menu, and its access level decides who sees it.
  • Included from the start. Access control, logging and the audit trail come with every screen.
  • Safe on production. If a Sketch file is not correct, the Builder refuses to build and asks for changes.
Built for regulated workEvery screen it builds comes with access control and an audit trail.
gifts.sketch
@sketch 1
@record
@table gifts
@width 900

The gift:
'Received by' fullname*: text | 'Email' email: text
'From' giver*: text | 'Value' amount*: money
'What it was' description* [textarea]: long

The decision:
'Manager' manager: text
'Decision' decision [select: pending, kept, returned, donated]: code
Gifts register

Ticketing

All the work in one queue. Nothing lost in an inbox.

Requests from staff, tasks for the team and queries from customers are collected in one place. People and AI co-workers work from the same list.

  • Email becomes a ticket. Mail to your service address is turned into a ticket.
  • Automatic filing. A classifier chooses the category and reports its confidence.
  • Defined actions. A ticket's status changes only through the actions you set, so the process is followed.
  • Due clocks. Due dates, an overdue filter and a service dashboard.
  • Configurable. Types, statuses, priorities and sources are yours to set.
Built for regulated workA ticket moves only through the actions you define, and each one is recorded.

Knowledge

Your documents, ready to answer.

Upload the documents your people already rely on into RAG stores. RAG stands for retrieval-augmented generation: a store splits and indexes its documents so the right passages can be found by meaning. A Knowledge is a selection of RAG stores that you name and give to a co-worker. The co-worker retrieves the matching passages, answers from them and says where the answer came from, so a person can check it.

  • Upload and go. PDF, Word and other formats are split and indexed into a RAG store automatically.
  • Sources shown. Answers point to the document and the page.
  • Separate RAG stores. Keep HR, legal and operations documents in stores of their own.
  • Knowledge is a selection. A Knowledge names one RAG store, several, or all of them. You choose which Knowledge each co-worker uses.
  • Your storage. Each RAG store is kept in Chroma, Postgres or SQL Server.
  • Used in workflows. Retrieval from a Knowledge is a step any workflow can call.
Built for regulated workA co-worker reads only the knowledge you allow, and its answers show their source.

Co-workers

AI colleagues with a job description.

A co-worker is an AI assistant set up for one role. It has a name, an owner, a model, the knowledge it may read, the tools it may use and the guardrails it works within. It signs in as a user with an access level, so it can do what a person at that level can do and nothing more.

  • Chat. Staff ask in plain words. Conversations are kept.
  • One setup screen. Role, model, knowledge, tools and guardrails.
  • Guardrails. Block or redact what goes in and what comes out.
  • Sanitiser. Personal details are replaced with labels before text goes to an outside model, and restored in the answer.
  • Report this answer. One click turns a doubtful answer into a ticket for a person to look at.
Built for regulated workEvery co-worker has an owner, an access level and guardrails.

Workflow

Repeatable work, on rails.

Most work that repeats does not need an AI to think it through each time. A workflow sets out the steps once, as a text file you can read. It uses a model only where judgement is needed, and it stops for a person where a decision matters.

  • Building blocks. A co-worker, a document search, a classifier, a branch, an approval, a platform action, an email, a wait for a reply, or another workflow.
  • Approvals. The flow waits until a person at the right level decides.
  • Safe by rule. A workflow that emails what an AI wrote cannot be saved without an approval on every path.
  • Parallel and retry. Branches run side by side, and failed steps try again.
  • Triggers. Start a workflow from incoming mail or from a newly filed ticket.
Built for regulated workEmail written by AI is never sent without a person approving it.

Connectors

Two controlled ways in. Both under your access rules.

Outside assistants and your own applications reach the platform through MCP or the API. A connection is a signed-in user with an access level. A call above that level does not exist for it.

  • MCP. A frontier-model assistant can raise and update tickets, or build grids, forms and reports for you.
  • API. The same registered calls, for your own systems and scripts.
  • Connected apps. OAuth 2.1 sign-in and personal tokens, listed on one screen.
  • Revoke. Cut off any connection when you need to.
  • Email. Incoming mail feeds tickets and workflows. Outgoing mail drafted by AI is marked as such.
Built for regulated workEvery connection signs in, has an access level and can be revoked.

Models

Use the model that fits the job. Change it when a better one arrives.

  • Local models

    Open models run on your own hardware. Sensitive work never leaves your network.

  • Frontier model

    A frontier model for work that needs the strongest reasoning, with the sanitiser in front of it where you switch it on.

  • Chosen per co-worker

    Each co-worker names its model. Swapping it is one setting, and the change is recorded with a reason.

Also in the box

The parts every system needs. Already there.

  • Dashboards

    Tiles and charts over your tickets and registers.

  • Scheduling

    Timed jobs and watched folders.

  • Users and access

    Access levels, two-factor sign-in and sign-in history.

  • Audit trail

    Every change to audited data, with who, when, before and after.

  • Messaging

    Email in and out, with mail logs and file handling.

  • Classifier

    A small open model that picks a choice and reports its confidence.

  • Platform health

    More than twenty self-checks on one screen.

  • Kill switch

    Stops the connectors and the models from one place.

See it built in front of you. From a sentence to a screen.

In a demo we take a request in plain words and turn it into a working, access-controlled register while you watch.