Hiring and onboarding
Checks applications against the role, drafts the welcome email and books interviews in the calendar.
Reads and drafts emails on its own. Sending one waits for approval.
AI employees for your business
AI Workforce OS gives your business a team of AI assistants that draft emails, sort customer questions, prepare reports and keep routine work moving. You ask in plain words, and nothing is sent or deleted until a person says yes.
Try it on this page with sample data. Nothing to install, and no technical skills needed.
How it works
No code and no training course. If you can brief a colleague, you can brief an AI assistant. Pick a step to try it.
Illustration. Nothing is sent.
1 assistant on your team. Each one can be changed after it is added.
Your AI team
Four examples from the demo workspace, each with a job, the tools it needs and a clear line it will not cross without asking. In the demo their tools work with practice data. Pick a tool to see what it may do there.
Checks applications against the role, drafts the welcome email and books interviews in the calendar.
Reads and drafts emails on its own. Sending one waits for approval.
Keeps tickets up to date, sums up open pull requests and writes the daily standup note, so nobody has to.
Reads channels on its own. Posting a message waits for approval.
Pulls together what your own documents already say, points out what is missing and writes the summary into a new document.
Reads files and creates new documents. It has no way to share them.
Sorts the overnight queue, drafts replies from your support handbook and passes on anything it cannot answer.
Drafts every reply from your handbook. Each one waits for approval before it is sent.
See it work
Two short demos you can use now. Where one waits, time is sped up.
Simulated in your browser. Nothing is sent, and no account is needed.
Approvals
SimulatedReading and drafting happen on their own. Sending an email, posting a message or deleting something stops here and waits for a person to approve it.
Try this
Reads the new starter’s details and works out what to say.
Drafting happens on its own.
Sending leaves the business, so it waits here for a person.
In this demo it goes to a practice mailbox. Nothing leaves your browser.
Writes a short note of what it did.
Approval needed · outgoing email
Exactly what will be sent
Time is sped up here. A real request waits until its deadline.
Answers with sources
SimulatedAsk a question and it answers from your own documents, pointing to the page behind each part of the answer.
Try this
What happens when an agent wants to send an email?
It does not send it straight away. Outbound actions go to an approval queue, where a person reviews them before they run . The proposal gives that review to a manager or the CEO .
For illustration: in the product, the reply is a short sentence with the matching passages listed under it. The search is replayed here.
Safe by design
The AI team does the work. People keep the decisions, and there is a record of every one.
Sending an email, posting a message or deleting something waits for a person who is allowed to approve it.
Five ready-made roles, from owner to viewer, decide who can ask, who can approve and who can only look. You can make your own.
Every approval decision and every finished job is written down against the person responsible, in a record built so that a changed entry can be detected.
Answers point to the page they came from. When your documents do not cover a question, it says so instead of guessing.
Works with 7 AI providers, including OpenAI, Anthropic and Google Gemini. If one is busy, it moves on to the next.
Set a timetable in plain words, such as every weekday at 9am, and pause it whenever you like.
The five roles every workspace starts with. Owners and admins can create roles of their own.
| What a person can do | Owner | Admin | Manager | Employee | Viewer |
|---|---|---|---|---|---|
| Ask the AI team for work | Owner can ask the AI team for work | Admin can ask the AI team for work | Manager can ask the AI team for work | Employee can ask the AI team for work | Viewer cannot ask the AI team for work |
| See the work as it happens | Owner can see the work as it happens | Admin can see the work as it happens | Manager can see the work as it happens | Employee can see the work as it happens | Viewer can see the work as it happens |
| Approve what gets sent | Owner can approve what gets sent | Admin can approve what gets sent | Manager can approve what gets sent | Employee cannot approve what gets sent | Viewer cannot approve what gets sent |
| Change how the assistants work | Owner can change how the assistants work | Admin can change how the assistants work | Manager can change how the assistants work | Employee cannot change how the assistants work | Viewer cannot change how the assistants work |
| Invite people and set their roles | Owner can invite people and set their roles | Admin can invite people and set their roles | Manager cannot invite people and set their roles | Employee cannot invite people and set their roles | Viewer cannot invite people and set their roles |
Questions
Companies that want AI assistants to take on routine work, such as hiring and onboarding, customer emails, research summaries and project updates, while people keep the final say.
No. Sending an email, posting a message or deleting something always waits for a person who is allowed to approve it. If nobody answers before the deadline, the request expires and nothing is sent.
Reading, drafting and routine updates, such as adding a calendar event or updating a ticket, happen on their own. Sending an email, posting a message and deleting something wait for a person. Owners and admins can make more actions wait for approval.
No. You ask for work in plain words in Chat, typed or spoken. The one-off set-up, such as choosing which AI provider to use, is done once by whoever looks after your workspace.
It tells you. Answers come from the documents you add, with the page each point came from. When nothing in your documents covers the question, it says it cannot answer rather than guessing.
Yes. Out of the box it runs on a built-in sandbox model that gives practice answers, clearly labelled as such, and every connector works with practice data, so nothing real is sent.
The one you choose: OpenRouter, Groq, NVIDIA NIM, Google Gemini, AWS Bedrock, Anthropic and OpenAI. You set the order, and if one is busy or unavailable it moves on to the next.
Yes. Gmail, Google Calendar, Microsoft Outlook and the other connectors connect to your real accounts once an owner or admin sets them up. Until then they use practice data, so nothing real is sent. Your IT team will find the technical details on the page for IT teams.
Each person gets a role. An employee can hand work to the AI team but cannot approve what it sends, and a viewer can look at reports but change nothing. Owners and admins can create roles of their own to fit your team.
Create your workspace. Whoever creates it becomes the owner, can start with ready-made assistants and can invite everyone else afterwards.
See it for yourself
See how your AI team drafts the work, asks before anything is sent and shows where its answers come from. The demos run on sample data, so nothing real is sent.