hangar docs.
How Hangar works, what it needs and what stays on your hardware. Written for the IT and operations teams who will run it.
Hangar is in early access. There is no public installer yet: we install it with you. The full API and MCP reference goes to early access customers with their installation.
overview.
Hangar runs ready-to-work computers for AI agents on your own servers. Each agent gets a Windows desktop with your programs and saved logins, and works on it with screen, keyboard and mouse, the way a person does.
That lets an agent work in programs that have no API: an ERP, a warehouse system, a clinic calendar, a supplier portal. The programs stay exactly as they are.
What Hangar is
- Infrastructure that runs on hardware you control: your own server, a VM in your cloud account or a workstation.
- Desktops for agents, desktop images with your programs and saved logins, a vault for logins, recordings and a click-by-click log, and an optional reset to a clean copy.
- An API and MCP that connect your agents, and the tools to manage all of this.
What Hangar doesn’t do
- It doesn’t build automations. The agent and the process it runs are a separate layer, built by your team, an implementation partner or with us.
- It doesn’t include an AI model. Your agent uses a model from Appto, Claude, ChatGPT or another provider, and that provider bills the usage.
- It doesn’t host your desktops. They run on your hardware, inside your network.
- It isn’t open source. The code is not available for inspection.
Where Hangar sits
Four layers make an agent do real work. Hangar is one of them.
concepts.
Seven words cover most of Hangar. The rest of these docs use them in this sense.
Desktop
A Windows desktop that one agent uses for one task. It has a screen, a keyboard and a mouse, and the agent works on it the way a person does. Each task gets its own desktop, kept apart from the other desktops.
Desktop image
The starting point for a desktop: Windows with the programs you chose, their settings and saved logins. You build an image once, and every desktop made from it starts the same way. During setup we build your first image with you.
Pool
The desktops made from one image on one server. When an agent asks for a desktop, Hangar hands it a free one from the pool. A pool can hold as many desktops as your hardware runs.
Vault
Where logins are kept, on your server. Saved sessions can be part of the desktop image, and the agent signs in with them, so your team doesn’t hand passwords to the agent in a chat.
Recording and log
Hangar records every session on a desktop: a video of the screen and a click-by-click log of what the agent did. Both are kept on your server, for review and audit.
Reset
After a task, a desktop can go back to a clean copy of its image, so the next task starts from the same state as the last one. Reset is optional. The recording and the log are kept either way.
Connecting an agent
Agents reach Hangar through MCP or the API. Appto, Claude and ChatGPT connect through MCP. n8n, Make, Zapier, Power Automate, UiPath and your own code connect through the API or MCP. The agent’s model looks at the screen to decide what to do next, and your model provider bills that usage. More on the API and MCP
how a task flows.
A task comes from a person in Appto, Claude or ChatGPT, from an automation tool or from your own code. From there it goes through four steps on one desktop.
-
The agent asks for a desktop
Through the API or MCP, the agent asks Hangar for a desktop from a pool. Hangar hands it a free one, made from the image you chose for that pool.
starting -
It works on the desktop
The agent looks at the screen, then types and clicks in your programs. It signs in with the saved logins from the vault. Hangar records the session as it goes.
at work -
It reports back
The agent returns the result to whoever sent the task. If a case is unclear, the agent can leave it for a person.
reporting -
The desktop is free again
The desktop goes back to the pool for the next task. If you choose, it first resets to a clean copy of its image. The recording and the log stay on your server.
free
requirements.
Hangar runs on hardware you control. We don’t host your desktops.
Where it runs
- A Linux server in your office, building or data centre.
- A VM in your own cloud account.
- A workstation, such as a single machine for a first process.
How many desktops
How many desktops run at once depends on the processor and memory of the machine, and on what the programs in your desktop image need. We size this with you before installation. We suggest starting with one process on one server and adding desktops as you hand over more work.
What we go through with your IT
- The machine you plan to use, and who on your side manages it.
- The programs that go into the desktop image, and their licences.
- How your agent will reach Hangar on your network.
- Which logins go into the vault, and which saved sessions go into the image.
security model.
Hangar runs inside your network. What leaves it is what the agent’s model needs in order to work: images of the screen and the task.
What stays on your hardware
Everything Hangar keeps is stored on your server. If you stop using Hangar, it stays there.
- Desktops and your programs
- Desktop images
- Files the agent works with
- The vault with your logins
- Recordings and the log
Appto, Claude, ChatGPT or the provider behind your agent. You choose which one.
The model sees the screen
An agent decides what to do by looking at the screen. To do that, it sends images of the desktop to its model. Anything visible on the desktop during a task can reach the model provider, under that provider’s data terms.
Two habits keep this in check: choose a model provider whose terms fit your policies, and keep programs and data the agent doesn’t need off its desktop image.
Desktops are kept apart
Each task gets its own desktop, kept apart from the other desktops. With reset turned on, a desktop goes back to a clean copy of its image before the next task.
Controls in Enterprise
Enterprise adds company sign-in (SSO), rules for which agent may open which programs, a person approving chosen steps such as payments, audit log export, your own retention rules for recordings, a DPA and a security review with your IT. Compare plans
getting started during early access.
There is no public installer yet. We install Hangar with you, remotely, on your hardware.
Request access
Tell us about the process you want to hand to an agent and the programs it uses.
We check the setup with your IT
The machine, the network and the programs for the desktop image.
We install Hangar with you
We install it on your server and build your first desktop image together. The technical setup is a one-time $2,000. See pricing
You connect your agent
You get the API and MCP reference and connect Appto, Claude, ChatGPT, an automation tool or your own code. The automation itself is built by your team, an implementation partner or with us.
API and MCP.
Agents and tools talk to Hangar through an API and an MCP server. Both cover the same work: getting a desktop, working on it and giving it back.
What an agent can do
- Ask for a desktop from a pool, and give it back when the task is done.
- See the screen, type and click on that desktop.
- Sign in to programs with logins from the vault.
- Find the recording and log of a session.
- Reset a desktop to a clean copy of its image.
- Report the result to whoever sent the task.
How each tool connects
| Agent or tool | Connects through |
|---|---|
| Appto, Claude, ChatGPT | MCP |
| n8n, Make, Zapier, Power Automate, UiPath | API or MCP |
| Your own code | API or MCP |
Model usage is billed by the model provider, and each automation tool is billed by its vendor.
The reference
The reference isn’t public yet. Early access customers get it with their installation. Request access
FAQ for IT.
Where does Hangar run?
On hardware you control: a Linux server, a VM in your own cloud account or a workstation. We don’t host your desktops.
What runs on the desktops?
Windows, with the programs you put in the desktop image. The agent uses them through the screen, keyboard and mouse, so the programs need no API and no changes.
What data leaves our network?
Images of the screen and the task go to the model behind your agent, so it can decide what to click. The desktops, files, the vault, recordings and the log stay on your server.
How are logins stored?
In a vault on your server. Saved sessions can be part of the desktop image, and the agent signs in with them.
Where are recordings and logs kept, and for how long?
On your server. Your own retention rules for recordings are part of Enterprise. We go through the default with you during setup.
Are desktops kept apart from each other?
Yes. Each task gets its own desktop. With reset turned on, a desktop goes back to a clean copy of its image before the next task.
What network access does Hangar need?
It depends on where your agent and its model run. We go through the network setup with your IT before installation.
Can Hangar run without internet access?
Installation without internet access is part of Enterprise. The agent still has to reach its model.
Does it support company sign-in (SSO)?
Yes, in Enterprise, along with automatic user setup, roles and approvals for chosen steps.
Can we review the source code?
No. Hangar is not open source, and the code is not available for inspection. Enterprise includes a security review with your IT and documents for your security team.
How are updates handled?
Updates are included in both plans. We agree how they reach your server during setup.
What happens if we stop using Hangar?
The servers are yours, and the desktops, files, recordings and logs are kept on them. Hangar is billed monthly, and you can cancel any month.
A question this page doesn’t answer?
Contact us