OS47 by theagency47 · Available now, in use with theagency47 clients

OS47: see, approve and test everything your AI agents do.

OS47 is theagency47's client workspace for AI agents: a private web app where your team sees what your agents know and do, approves every reply and knowledge change before it is sent, checks each agent against test questions with known answers, and can stop every agent at once. It is included at no separate price with every theagency47 product: Spark, Workforce, the AI Sales Agent and the AI Company Brain. It runs in theagency47's Cloudflare account or in your own.

We reply within one business day with a calendar link.

Illustration of the OS47 approval screen: sending on its own is off and locked by 47
Sending on its own is off and locked for every agent. Illustration based on the OS47 workspace.
  • Nothing goes out without your OK
  • Comes with every agent we build
  • From one Spark agent to the AI Company Brain
  • Stop every agent at once

Business data stored in the EU · The rules behind these screens →

What is OS47?

OS47 is theagency47's client workspace for AI agents: a private web app where your team sees what your agents know and do, approves every reply and knowledge change before it is sent, checks each agent against test questions with known answers, and can stop every agent at once.

  • No separate price. OS47 is included with every theagency47 product: Spark, Workforce, the AI Sales Agent and the AI Company Brain.
  • Your choice of hosting. It runs in theagency47's Cloudflare account or in your own.
  • Available now. OS47 is in use today with theagency47 clients.
theagency47 · Updated October 2026
The workspace

Everything your agents do, on six screens

The illustrations below are based on the OS47 workspace of a guesthouse with two agents, Concierge and Ops.

Approvals

Agents propose; a person on your team approves. Every reply an agent drafts and every change to what your agents know waits here until someone decides. Nothing is sent on its own.

Illustration of the OS47 Approvals queue: agent replies waiting for a person to approve or reject
Agents propose, a person approves. Nothing is sent on its own. Illustration based on the OS47 workspace.

Trust

The Trust screen holds the kill switch: one button pauses every agent at once. You are asked twice before it acts, and nothing is deleted.

Illustration of the OS47 Trust screen: one kill switch that stops every agent at once
One button pauses every agent. Illustration based on the OS47 workspace.

Quality

Each agent is checked against test questions with known answers, an eval suite. The Quality screen shows each check, when it ran and the agent's score against a 90% target. A change goes live only when it meets or beats the target.

Illustration of the OS47 Quality screen: each agent's score on test questions against a 90% target
Agents are tested against questions with known answers. Illustration based on the OS47 workspace.

Knowledge

Knowledge lists what your agents answer from: your texts, prices, policies and answers to common questions. Anything not reviewed in 90 days is flagged as out of date, and you can request a fix from the same screen.

Illustration of the OS47 Knowledge screen: the sources agents answer from, one flagged as out of date
What your agents answer from. Anything not reviewed in 90 days is flagged. Illustration based on the OS47 workspace.

Activity

Activity shows, day by day and per agent, the conversations handled, tasks done and completed, what was handed to a person, and the time saved.

Illustration of the OS47 Activity screen: conversations, tasks and hand-offs per agent, day by day
What each agent did, day by day. These numbers are not client results. Illustration based on the OS47 workspace.

Home

Home is the overview: each agent's status, and in plain words what your agents may and may not do.

Illustration of the OS47 Home screen: what the agents may and may not do, with sending on its own locked
What every agent may and may not do, on the home screen. Illustration based on the OS47 workspace.

See these screens with your own agents on a call →

Human in the loop

How an approval works, step by step

  1. An agent drafts a reply, a task or a change to what it knows.
  2. The draft appears in Approvals, waiting for your decision.
  3. Someone on your team reads it and edits the text if needed.
  4. They approve or reject it. Nothing goes out before that.
  5. OS47 records who decided and when.

This is human-in-the-loop in practice: the agent does the work, a person makes the call. How we secure agents beyond approvals is in our AI agent security guide.

Illustration of an agent's draft in OS47 waiting for a decision, with text a team member can edit
Every reply can be edited before approval and is recorded. Illustration based on the OS47 workspace.
Per agent

Controls for each agent

Each agent has its own settings. Your team chooses its tone (warm, neutral or formal), its working hours and the signature added to its messages. Sending on its own is off and locked: only theagency47 can change it, and only after you agree.

Illustration of the OS47 controls for one agent: tone (warm, neutral or formal) and working hours
Each agent's tone and working hours. Illustration based on the OS47 workspace.

Switch off for good

When you no longer need an agent, you can switch it off for good from the same page.

Illustration of the Switch off for good control on an agent's settings page in OS47
Switch an agent off for good from its settings page. Illustration based on the OS47 workspace.
Every product

Which agents show up in OS47

Every agent theagency47 builds for you shows up in OS47, at no separate price.

Launch, Run and Scale package prices → Pricing

Knowledge

OS47 and the AI Company Brain

The AI Company Brain is the knowledge your agents answer from. OS47 is where your team sees that knowledge and steers it: the Knowledge screen lists every source and flags anything out of date, and a change to the knowledge goes through Approvals just like a reply.

If your company knowledge lives in Obsidian, the Second Brain is the vault your agents read from and write back to.

Hosting

Where OS47 runs: theagency47's Cloudflare account or yours

You choose where OS47 runs. Both options run on Cloudflare, and both are set up for your business only. theagency47 builds and runs AI agents on Cloudflare.

In theagency47's Cloudflare account In your own Cloudflare account
Who holds the account theagency47, with a setup used by your business only Your company, in your name
Who runs it day to day theagency47 theagency47, with access you grant
Hosting bills Included in your theagency47 plan Billed to you directly by Cloudflare
Fits Teams that want nothing to manage Companies whose IT or procurement want the account in their name

Where your data lives. Your business data is stored in the EU. Model processing may run outside the EU.

How we handle data →

Example

Example: a guesthouse running two agents

The illustrations on this page are based on a small guesthouse with two agents.

  • Concierge drafts replies to guests, such as a question about bringing a pet. Each reply waits in Approvals until someone at the guesthouse approves it.
  • Ops works for the team: it drafts the weekly housekeeping summary, which waits for approval in the same queue.

See what agents do for hotels, villas and short-term rentals: AI agents for hotels and villas.

Illustration of a Concierge reply to a guest about pets in OS47, approved by a teammate, with both times recorded
A guest reply drafted by Concierge and approved by a teammate. Illustration based on the OS47 workspace.

Questions about OS47

What is OS47?

OS47 is theagency47's client workspace for AI agents: a private web app where your team sees what your agents know and do, approves every reply and knowledge change before it is sent, checks each agent against test questions with known answers, and can stop every agent at once.

Is OS47 built on Cloudflare OS, or made by Cloudflare?

No. OS47 is built by theagency47 and is not built on Cloudflare OS, Cloudflare's open-source agent workspace, which theagency47 does not offer. OS47 runs on Cloudflare, in theagency47's account or in yours. theagency47 is an independent agency, not a Cloudflare partner, and is not affiliated with or endorsed by Cloudflare.

How much does OS47 cost?

OS47 has no separate price. It is included with every theagency47 product: Spark, Workforce, the AI Sales Agent and the AI Company Brain.

Where does OS47 run, and where is my data?

You choose: OS47 runs in theagency47's Cloudflare account, set up for your business only, or in your own Cloudflare account. Your business data is stored in the EU. Model processing may run outside the EU.

Is OS47 available now?

Yes. OS47 is in use today with theagency47 clients.

Can an agent send a message without approval?

Not by default. Sending on its own is off and locked for every agent, so each reply waits for a person on your team. Only theagency47 can change this, and only after you agree.

How do I know the agents answer correctly?

OS47 tests each agent against questions with known answers and shows the score against a 90% target.

Can I stop all agents at once?

Yes. The Trust screen has a kill switch that stops every agent at once.

Weighing whether to build your agents yourself? Read DIY AI agents with Cursor & Grok Bot vs OS47.

See OS47 on a call

Bring one process; we show the six screens with it in mind.

We reply within one business day with a calendar link.

Updated October 2026. The images on this page are illustrations based on the OS47 workspace.

Cloudflare and Cloudflare OS are trademarks of Cloudflare, Inc. theagency47 is an independent agency and is not affiliated with or endorsed by Cloudflare.

Part of