# Ghost AI

> Source: https://aiwiki.ai/wiki/ghost_ai
> Updated: 2026-10-15
> Categories: AI Agents, AI Companies, Consumer AI Devices
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Ghost AI." aiwiki.ai, 15 Oct 2026. https://aiwiki.ai/wiki/ghost_ai
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

Ghost AI, branded as Ghost, is a company developing personal computers for continuously running [AI agents](https://aiwiki.ai/wiki/ai_agents). Its first product, Core, is a dedicated home computer intended to run models locally and retain personal context across interactions. Ghost describes Core as a system that connects to applications, files and devices rather than an assistant used only through individual prompts.[1][2]

## Company and launch

Founder and chief executive Zain Javaid announced an $11 million financing round in October 2026. Reporting on October 5 identified Andreessen Horowitz as the lead investor, with Abstract, Audacious Ventures, Nova and SV Angel participating.[3] The company's privacy policy identifies its legal operator as ZMJ, Inc., doing business as Ghost.[5]

Ghost's stated approach is to keep personal intelligence and its accumulated memory on hardware owned by the user. Its company description also discusses an intended ecosystem of devices for capturing context and accessing an agent, including cameras and wearables; these plans should not be read as a list of separately released Ghost products.[2]

## Core

As of October 16, 2026, Core's product page listed a $3,499 price, marked the first batch sold out, and accepted registrations of interest for batch 2. Batch 1 was scheduled to ship on October 31, a future date at the time of that listing.[4]

### Hardware

Core combines a workstation GPU with a desktop processor, system memory and local storage. The GPU uses [NVIDIA Blackwell](https://aiwiki.ai/wiki/nvidia_blackwell) architecture.[1][6]

| Component | Published specification |
|---|---|
| GPU | NVIDIA RTX PRO 4000 Blackwell SFF Edition [6] |
| GPU memory | 24 GB GDDR7 with error-correcting code (ECC) [6] |
| GPU memory bandwidth | 432 GB/sec [6] |
| GPU AI performance | 770 TOPS [6] |
| CPU | AMD Ryzen 5 7600, 6 cores, 12 threads [1] |
| System memory | 64 GB DDR5 [1] |
| Storage | 1 TB NVMe SSD [1] |
| Wireless connectivity | Wi-Fi and Bluetooth [4] |
| Dimensions | 72 mm wide, 380 mm high, 220 mm deep [4] |
| Weight | 4.6 kg [4] |
| Enclosure | Stainless steel, brushed silver finish, acrylic light diffuser [4] |

NVIDIA defines the GPU's 770 TOPS figure as theoretical FP4 throughput using sparsity, with peak rates based on GPU Boost Clock. It is a hardware specification, not a measured response-time result for Core's assistant.[6]

### Software and access

Ghost describes Core as an always-on system that accumulates context from connected sources and can initiate suggestions without a new prompt. Its demonstrations include information from email, calendars, files, browser activity, fitness wearables and smart-home devices. These are the company's descriptions of the product's intended operation.[1]

The setup guide describes pairing through the Ghost app, joining Wi-Fi, selecting a model, connecting data sources and building memory. It also documents downloading a different model through Settings > Model while preserving existing context. Core starts when power is connected and has no power button.[7]

Ghost lists desktop access from Mac, Linux and Windows, and phone access from iPhone and Android. Its FAQ says Core provides an OpenAI Responses API-compatible endpoint for applications that support custom endpoints, including [opencode](https://aiwiki.ai/wiki/opencode), [OpenClaw](https://aiwiki.ai/wiki/openclaw) and [OpenAI Codex](https://aiwiki.ai/wiki/openai_codex). Ghost advertises local inference without subscription or per-token inference fees.[4]

## Local processing and online services

Ghost's October 5, 2026 privacy policy says conversations, files, connected-app content and memories are processed and stored locally, without remote-model fallback for conversational AI or memory. Local processing does not mean that every feature operates offline.[5]

For services such as web search and phone calls, requests can pass through Ghost's gateway to third-party providers. Ghost says it does not retain forwarded content, but providers apply their own retention policies. Users can configure their own provider keys. Remote access can use Ghost's encrypted relay or a separate connection such as Tailscale. Ghost still handles operational, purchase and support records.[5]

Screen history is disabled initially. When enabled, it captures images and recognized text; private browsing does not pause capture. Memories can survive deletion of their source, and paired devices may retain separate copies. The policy invites independent security reviews, rather than presenting a completed audit. These are Ghost's published data-handling commitments, not an independent certification.[5]

The product FAQ separately notes that internet access is needed for web search, email and software updates even though the AI and memory run locally.[4]

## References

1. Ghost. [Ghost AI: Core overview](https://ghost.ai/). Product description and component specifications. Accessed October 16, 2026.
2. Ghost. [About](https://ghost.ai/about). Company description and product approach. Accessed October 16, 2026.
3. Mike Wheatley. [AI agent hardware startup Ghost, led by its 19-year-old-founder, raises $11M](https://siliconangle.com/2026/10/05/ai-agent-hardware-startup-ghost-led-by-its-19-year-old-founder-raises-11m/). SiliconANGLE, October 5, 2026.
4. Ghost. [Core](https://ghost.ai/core). Product page, expanded specifications and frequently asked questions. Accessed October 16, 2026.
5. Ghost. [Ghost and Core Privacy Policy](https://ghost.ai/privacy). Effective October 5, 2026. Accessed October 16, 2026.
6. NVIDIA. [NVIDIA RTX PRO 4000 Blackwell SFF Edition](https://www.nvidia.com/en-gb/products/workstations/professional-desktop-gpus/rtx-pro-4000-sff/). Specifications and performance footnotes. Accessed October 16, 2026.
7. Ghost. [Setup Guide](https://ghost.ai/setup). Updated September 18, 2026. Accessed October 16, 2026.