MangoTart is a dedicated computer for running a personal AI agent. It arrives with MangoTart Agent OS, Hermes, a browser runtime, a local management console, updates, diagnostics, and backup/recovery controls already installed. It is designed for people who want an always-on agent without turning a laptop into a server or maintaining a DIY stack.
It is not a model subscription, a local-LLM workstation, or a promise that an autonomous agent will always make the right decision. Read this page before ordering so the boundaries are clear.
MangoTart is a good fit if
- You want one personal agent that can keep running after your laptop closes.
- You want to use your own ChatGPT subscription, API key, OAuth account, or compatible endpoint.
- You value a ready browser runtime, local console, automatic updates, diagnostics, and a supported backup path.
- You are comfortable connecting selected accounts and services to an agent and reviewing sensitive actions.
- You have a normal home or small-office network with an Ethernet port available for first setup.
It may not be the right fit if
- You want to run a large frontier model fully offline on the device. MangoTart runs the agent runtime locally, while model requests normally go to the provider you choose.
- You need a screen, keyboard, or desktop computer. MangoTart is headless and is managed from another phone or computer.
- You need the management console to be publicly reachable while away from home. The console is currently LAN-only.
- You need several agents or separate private profiles for family or team members on one device. One MangoTart is one trust boundary; its agent memory and connected accounts are not separated per person.
- You specifically need OpenClaw. MangoTart currently ships Hermes as its agent runtime.
- You need guaranteed outcomes from autonomous work. The agent can misunderstand instructions, encounter changed websites, or be limited by providers and connected services.
What you need
| Requirement | Why |
|---|---|
| Internet connection | Model providers, messaging platforms, websites, updates, connected tools, and optional cloud backup use the internet. |
| Router or network with Ethernet | First setup uses a wired connection. Wi-Fi can be configured after onboarding. |
| Phone or computer on the same LAN | You open the local management console from your own browser. |
| Email address | Your MangoTart account uses one-time email codes and binds the device for ownership, warranty, backup access, and support. |
| Model access | Bring a supported ChatGPT subscription, API key, OAuth account, or compatible endpoint. See Models and ongoing costs. |
| One chat app | Telegram is the quickest common starting point, but availability depends on your region. Other supported channels can be added later. |
First setup is wired
MangoTart does not use a captive portal, Bluetooth setup, or a temporary Wi-Fi hotspot. Connect the included Ethernet cable for onboarding, then switch to Wi-Fi from the management console if you want. See Network and access.
What arrives
The package includes:
- the assembled MangoTart agent computer with Agent OS installed;
- an official 27 W USB-C power supply with the plug type shown for your storefront currency;
- a 1 m Ethernet cable; and
- a printed quick-start guide.
It does not include a display, keyboard, mouse, model subscription, API credit, mobile data plan, or paid third-party service account.
What stays local — and what does not
Hermes state, memory, skills, configuration, credentials, chat state, and the browser profile live on the device. That does not make the complete workflow offline: prompts and relevant context go to your model provider; messages pass through the messaging platform; browser activity reaches the websites you visit; and external tools receive the data needed for their task.
Provisioned devices also use MangoTart services for registration, heartbeat, updates, support, and optional encrypted backup. Authorized support operators have an administrative remote-maintenance path that does not currently require per-session approval in the product UI. Read Privacy, security, and permissions and the complete data flow before connecting sensitive accounts.
One device is one trust boundary
Several people can technically message the same agent, but they share its memory, tools, credentials, and connected accounts. MangoTart does not provide separate private workspaces for each family or team member on one device.
If two people must not see or influence each other's agent context, they need separate devices. Group chats can create separate conversations, but they are not a security boundary for the underlying agent state.
Access while you are away
The management console is reachable only on the same local network. You cannot currently open it directly from the public internet. However, the agent computer can keep running while you are away, and you can continue talking through a configured chat platform. Scheduled work can also continue if power, internet, the model provider, and the delivery channel remain available.
Costs after purchase
MangoTart itself is a one-time hardware purchase and currently has no required MangoTart subscription. You remain responsible for:
- your ChatGPT subscription or usage-based model-provider bill;
- paid tools and services you connect, such as Twilio or premium APIs;
- normal internet and electricity costs; and
- applicable customs duties, import taxes, GST/VAT, or similar charges.
See Models and ongoing costs and Orders, shipping, returns, and warranty for the details.
Before ordering checklist
- I can connect the device to Ethernet for first setup.
- I understand that the console is LAN-only.
- I have, or plan to obtain, compatible model access.
- I understand that model requests and connected services still receive relevant data.
- I am comfortable with one shared trust boundary per device.
- I have read the shipping, tax, cancellation, and warranty terms for my order.
If those points work for you, continue with the Quick start when your MangoTart arrives.