Ghost Raises $11 Million for a $3,499 Computer Built to Run Personal AI

A personal AI agent that keeps working without constant instructions is Ghost’s pitch for a $3,499 computer called Core. The startup emerged from stealth on October 5, 2026, pairing its first product with an $11 million seed round led by Andreessen Horowitz, TechCrunch reported. Its promise: keep models and personal memory on hardware the customer owns.

Ghost spent a year building before the announcement. Abstract, Audacious Ventures, SV Angel and Nova also joined the financing. Preorders began October 5, with shipping scheduled for the last week of October. SiliconANGLE subsequently reported that the first batch had sold out and Ghost was collecting interest in a second batch, without a date for more orders.

A computer without a screen

Core is not a desktop users sit in front of. It has no screen; owners access it through a phone or app. Voice input also goes through the Core app, rather than directly into the box. CEO Zain Javaid, 19, describes the machine as a dedicated home for agents—AI systems designed to take actions on someone’s behalf.

The price includes an Nvidia RTX Pro 4000 SFF Blackwell GPU, a chip used to run AI models. Javaid says Core also supplies its own software layer, browser and file system. Three models come preinstalled, and owners can add models from Hugging Face, a model-sharing platform, or load models they have trained themselves.

Asked how that differs from a Mac mini configured for AI, Javaid pointed to the bundled setup: Core’s hardware and software are built for agents, with models already installed. Ghost can deliver model updates over the air. Users can choose whether to switch to a new model generation or enable automatic updates.

Memory that feeds an always-running agent

Javaid’s goal is to bring information from a user’s desktop, apps and smart-home devices into one machine. He says Core can continuously process information and tasks without repeated prompting. The intended assistant would build an understanding of a person’s life, rather than simply respond to each new request in isolation.

His example is a notification suggesting someone might be getting sick. The agent would combine an elevated resting heart rate, computer history showing late-night work, and local flu-season context. That is Javaid’s illustration of the intended behavior, not a demonstrated health result. He says owners can control how much autonomy their agents receive.

All of the source code, logic, and model weights live on the device, not the services.

Zain Javaid, Ghost CEO, speaking to TechCrunch

Ownership, with safeguards still being built

Javaid says models and personal memory run locally, with personal information encrypted so Ghost and other third parties cannot meaningfully access it. He also says Core is designed to keep working if Ghost shuts down. The startup’s continuity promise rests on keeping the code and models on the purchased machine, rather than requiring an outside operator to maintain a service.

Control over credentials is another part of the design. Javaid told TechCrunch the team is building a firewall to limit how agents use logins and passwords. It is intended to monitor outgoing network requests, reject improper data transfers and flag them for user review. He described that protection as work being built, not a completed safeguard.