Autonomous.ai is selling a $299 pyramid-shaped computer called Intern 2 for running AI agents such as Hermes and OpenClaw away from your main laptop. The interesting part is not the glowing capstone. It is the attempt to make unsupervised agent work feel like an appliance rather than a software experiment with access to everything you own.
Watch Desk analysis
What happened
The Register reports that Intern 2 is preconfigured with Hermes or OpenClaw, with the option to install another agent for an extra $50. The 4.7-inch device includes a microphone, speaker, USB-C, Wi-Fi 5 and Bluetooth 5, and uses cloud inference rather than running a model locally.
Autonomous.ai says the machine receives an allowance of ten million tokens a day through its own gateway. Dee Tran, the company’s head of AI hardware, told The Register that an internal router selects between models for each task. The device can also connect to a local model running on separate hardware through an OpenAI-compatible server.
The company’s product information is not perfectly tidy. The listing names an OrangePi board, while the press release says the machine can include a Raspberry Pi 4 or 5. Autonomous.ai had not answered The Register’s requests for clarification when the article was published.
Why it matters
Intern 2 is a small but telling answer to a growing problem: people may want agents to handle cron jobs, email, scheduling, GitHub work or research, but may not want those agents roaming around their everyday computer. A separate box creates a clearer boundary, even if it does not magically make an agent trustworthy.
The trade-off is equally clear. Cloud inference means users are buying a managed service as well as a piece of hardware, and the ten-million-token allowance is not the same thing as unlimited autonomy. The machine is a neat container for agent work, not a guarantee that the work will be sensible.
Our read
This is one of the more credible signs that agent computing is becoming a product category rather than merely a feature buried in a developer tool. The pyramid is memorable, but the real pitch is separation: put the agent somewhere it can be watched, configured and switched off without taking your laptop, inbox and personal files along for the ride.
That makes Intern 2 worth watching, while leaving the sensible buyer’s checklist intact. Ask which board is actually inside, where prompts and task data go, what the gateway logs, and how permissions are handled before giving an agent the keys to your digital house. A glowing pyramid is still a computer with opinions.
What to watch
- Hardware clarification
Autonomous.ai needs to resolve whether Intern 2 ships with OrangePi hardware or a Raspberry Pi variant. - Cloud boundaries
The gateway’s model routing, logging and data-retention terms matter as much as the box. - Agent permissions
Real-world safety will depend on how email, files, GitHub and social accounts are isolated. - Local-model support
A practical local inference path could make the device more private and less dependent on the company’s service.
Discussion spark: Would you trust a dedicated appliance with background agent tasks more than an agent running on your main computer, or does cloud inference simply move the risk elsewhere?
Sources and evidence
- If a Mac mini is agentic overkill, try this glowing pyramid that runs personal bots – The Register (16 September 2026, 23:28 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.