A 19-year-old's startup sells a screenless home computer built to run AI agents
Ghost, a San Francisco startup, launched Core, a home box that runs AI locally instead of in a data center, and a reader can judge the idea by its specs.
The short version
- Ghost raised eleven million dollars in seed funding from investors including Andreessen Horowitz.
- Core has no screen, mouse or keyboard; you control it through a smartphone app.
- Inside: AMD Ryzen 5 7600, Nvidia RTX Pro 4000 SFF Blackwell GPU, 64 gigabytes of RAM.
- Open models such as Qwen and Gemma come preinstalled; more can be downloaded from Hugging Face.
- A built-in firewall checks every outgoing request from the agents to stop data leaking out.
Why it matters
Most AI today runs in someone else's data center, which means your data travels and the service can change or disappear. Core is one concrete attempt to flip that, so you can ask what local AI really costs and what it can do. Treat it as a promise until independent speed tests exist.
A teenager's bet against the data center
While OpenAI and other big AI providers put their money into data centers, a much smaller company is betting on the opposite. Ghost, based in San Francisco, was founded by 19-year-old Zain Javaid. As t3n reported, drawing on TechCrunch, the startup raised eleven million dollars in seed funding from investors that include the venture capital firm Andreessen Horowitz.
Javaid's background is in financial mathematics, where he made investment decisions based on data. He then switched to building hardware for running AI on your own premises. The first product is called Core and is priced at $3.499 (German notation for three thousand four hundred ninety-nine dollars).
A computer with no screen
Core ships without a display, mouse or keyboard. It is meant to sit quietly in the background and run AI agents, which are programs that carry out tasks on their own. According to the report, these agents continuously collect data from the user's surroundings and work on tasks based on it.
You talk to the device through a smartphone app. As SiliconANGLE explained, that app handles everything, including voice input, which is not done on the device itself. Through it, buyers give the agents tasks and communicate with them.
What is inside the box
Running language models locally takes serious hardware, and the specifications on Ghost's website reflect that. They list the following parts:
- An AMD Ryzen 5 7600 processor with six cores.
- An Nvidia RTX Pro 4000 SFF Blackwell graphics processor, with a memory bandwidth of 432 gigabytes per second. Bandwidth matters because it limits how fast a model can read its own data while writing answers.
- 64 gigabytes of DDR5 working memory.
- A solid-state drive of one terabyte for data and installed models.
The case is brushed stainless steel with an acrylic light diffuser. These are the maker's own figures. Nothing in the report shows how fast the machine actually answers, which is the number a buyer would want to know.
Privacy by keeping everything at home
The main selling point is that nothing goes to the cloud. All calculations happen on the device, an approach called edge computing, meaning the work is done at the edge of the network, close to the user, rather than in a distant data center. Ghost says this keeps sensitive information such as health readings or smart-home data inside the house.
The startup also says locally stored data is encrypted so that Ghost itself cannot read it. On top of that, Core includes a firewall aimed at the agents' behavior. It checks every outgoing request the software sends to the outside network.
The goal is to prevent data exfiltration, the unwanted leaking of things like passwords or private documents. The report calls this a basic security risk for autonomous systems. Whether the firewall catches everything is untested in the source.
Open models, and a price that needs a reason
Core is not tied to one closed AI model. It comes with open models preinstalled, including versions of Qwen and Gemma, and its operating system can download more from platforms such as Hugging Face. Users can therefore swap models to suit their taste. Software updates and new model generations are meant to arrive automatically over the network.
Because the price is high, independence is part of the pitch. Founder Javaid told TechCrunch, in the report's paraphrase, that the device is designed not to depend on an outside party keeping a service running. That addresses a real worry with cloud AI, where a provider can change terms or shut down.
Still, a young company's first product is a promise until others test it. A fair question for any buyer is how fast it runs a named model, and what happens to the machine if Ghost one day disappears.
What to watch
- Do independent tests show how fast Core runs real models on its stated hardware?
- Does the agent firewall hold up when outside researchers try to get data out?
- Can a startup this young keep updates and new model support coming for buyers over time?
Try it
Before considering any home AI box, run an open model on a computer you already own and note how fast it answers. That tells you what speed you actually need.
Sources and links
- Reportingt3n: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by t3n · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
