Agent computing leaves the cloud, a $11 million bet on independent personal processors

Ghost has emerged from the secret-building phase with a screenless desktop device, putting an initial funding of $11 million behind a unusual bet: removing smart agents from the grip of cloud servers and returning them to independent physical hardware that sits on the user’s desk. The funding round was led by Andreessen Horowitz, with participation from Abstract, Odysseus Ventures, SV Angel and Nova, to develop the Core device, priced at $3,499, with shipments slated for the last week of October.

The company was founded by Zain Javed, 19, after working in mathematics- and data-driven quantitative trading and becoming bored, until the launch of ChatGPT in late 2022 drove him to explore personal AI. The founding team in the San Francisco facility includes Nicholas Chua, 19, Yifei Chen, 24, and Gautam Charda, 23. The device’s concept is to gather the user’s data from their apps, devices and computers into a single piece of hardware, serving asBrain inside a boxExecutes tasks and acts on behalf of its owner without the need for continuous textual commands.

The device is built on an architecture that differs fundamentally from general-purpose computers that are repurposed for smart tools; it includes a NVIDIA RTX Pro 4000 SFF Blackwell GPU integrated into the retail price without requiring a separate purchase. The device runs on a proprietary software layer with an integrated browser and file system, accessible via a phone app or voice interaction through the same app. It ships with three pre-installed models: Quinn-3.8-Next, Quinn-3.8-27B, and Gemini-4-31B, with the option to install additional models from the Hugging Face platform or load locally trained models. The system also provides over-the-air updates reminiscent of Tesla’s approach, allowing the user to choose between moving to newer generations or using automatic updates.

The most compelling engineering value lies in independence and sovereignty; the device is programmed to keep operating even if Ghost shuts down, because the model weights, software logic and code reside in the hardware itself rather than in external cloud services. All data remain encrypted locally, with no cloud provider or third party, including the manufacturer, able to inspect it. The system is completed by a local firewall built by the team to monitor all outbound network requests from the device, rejecting any request that attempts to exfiltrate data or exceed permissions and routing it for review. This architecture allows the agent to take proactive measures freely, such as detecting an elevated heart rate combined with late-night work and a seasonal flu outbreak in the city, and then notifying the user of a possible fatigue.

This experiment imposes a highly significant shift for consulting firms, banks and family businesses in the Gulf and Egypt; we have long observed hesitancy in handing over records and decisions to agents running on foreign servers that lack strict sovereignty guarantees. Owning an independent, network-isolated processing unit gives a decision-maker in Riyadh or Dubai a path to use agents without risking identity or credential leakage. Also calculate the difference in development centers in Cairo and Amman, where the accumulated cloud inference cost turns into a fixed capital asset, freeing the developer from dollar-fluctuating API bill volatility.

We see in this step an indication thatThe battle of independent agentsStarted leaving the clouds of major companies to settle in personal hardware and endpoint firewalls. The skill required from technical teams in the region no longer limits to writing cloud commands, but extends to managing model weights locally and auditing outbound traffic to protect permissions. And when data confidentiality becomes non-negotiable, the gap between renting a remote agent and owning its processor outright becomes the same gap between protecting your business and absolute technical dependence.