# Greyhaven

> Greyhaven builds custom sovereign AI systems: on-premise inference, private data pipelines, and model-agnostic architecture so enterprises maintain full control over their AI.

Greyhaven is an AI infrastructure company focused on data sovereignty, process-level security, and enterprise AI deployment. Core products include greywall (process sandboxing for AI workloads) and greyproxy (network-layer traffic control and allowlisting). Greyhaven's framework defines AI sovereignty through three pillars: Choice (model selection and routing), Control (data exposure and access boundaries), and Clarity (full observability into AI system behavior).

## Products

- **Greywall** ([greywall.io](https://greywall.io)): A kernel-level security sandbox for AI agents. Wraps AI coding agents in a container-free, multi-layer enforcement layer using Seccomp, Landlock, eBPF, namespaces, and a network proxy. Every file access, network request, and subprocess is intercepted before it executes. Open source under Apache 2.0.
- **Greyproxy** ([github.com/GreyhavenHQ/greyproxy](https://github.com/GreyhavenHQ/greyproxy)): Network-layer traffic control and allowlisting for AI workloads. Enforces outbound network policy at the process level, giving operators visibility and control over which hosts an AI agent can reach.

## Pages

- [Home](https://greyhaven.co/): Overview of Greyhaven's sovereign AI systems offering.
- [Sovereign AI](https://greyhaven.co/sovereign-ai): Definition and framework for AI sovereignty as an infrastructure property.
- [Insights](https://greyhaven.co/insights): Articles, case studies, devlogs, and media from the Greyhaven team.
- [Contact](https://greyhaven.co/contact): Get in touch with the Greyhaven team.
- [Demos](https://greyhaven.co/demos): Live demonstrations of Greyhaven products.

## Insights

- [How Greywall Prevents Every Stage of the LiteLLM PyPI Supply Chain Attack](https://greyhaven.co/insights/how-greywall-prevents-every-stage-of-the-litellm-pypi-supply-chain-attack): A 3-stage supply chain attack on litellm harvests SSH keys, cloud credentials, and Kubernetes configs — greywall's default-deny sandboxing stops it at every stage without any malware-specific rules.
- [What Actually Happens Under the Hood When You Prompt an AI Coding Agent](https://greyhaven.co/insights/what-actually-happens-under-the-hood-when-you-prompt-an-ai-coding-agent): AI coding agents make more outbound requests than you'd expect — network-layer tools like greywall reveal what's actually happening behind the scenes.
- [Why We Built Our Own Sandboxing System](https://greyhaven.co/insights/why-we-built-our-own-sandboxing-system): Don't let your AI vendor audit their own security. Greyhaven built an independent sandbox because their incentive is "just works", not "locked down".
- [Greyhaven Sovereign AI Framework](https://greyhaven.co/insights/greyhaven-sovereign-ai-framework): Many enterprises believe they're executing an AI strategy — but Choice, Control, and Clarity at the infrastructure level is what separates sovereign AI from vendor dependency.
- [Reflector: Infrastructure for Meetings, by Greyhaven](https://greyhaven.co/insights/reflector-infrastructure-for-meetings-by-greyhaven): Most organizations run on meetings — strategy, decisions, and context all live there. Reflector captures and structures that organizational knowledge automatically.
- [Case Study: AI-Driven Planning System for Case Workers](https://greyhaven.co/insights/case-study-ai-driven-planning-system-for-case-workers): A US-based non-profit engaged Greyhaven to build an AI planning product designed to improve social mobility outcomes for public-sector case workers.
- [Case Study: Sovereign AI in Enterprise Software Modernization](https://greyhaven.co/insights/case-study-sovereign-ai-in-enterprise-software-modernization): A major telecommunications provider modernized millions of lines of legacy COBOL using Greyhaven's sovereign AI infrastructure.
- [Max and Corey Discuss AI](https://greyhaven.co/insights/max-and-corey-discuss-ai): AI security, the future of software development, benchmarks, verification, and more.