Greyhaven
Greyhaven builds sovereign AI systems and private, self-hosted AI products that run inside customers' environments. Its approach emphasizes client-side data, auditable permissions, open-source foundations, secure-by-default design and embedded engineering delivery.
At a Glance
- SME teams whose decisions rely on institutional knowledge
- Public-sector agencies and regulated environments where data must stay within the perimeter
- Research and technical organizations with repeated analysis workflows
- Infrastructure and service providers needing private internal tooling
- +2 more
AI Tools by Greyhaven
(1)Greywall
Sandbox for AI Coding Agents
Discussions
No discussions yet
Be the first to start a discussion about Greyhaven
Latest News
How Greywall Prevents Every Stage of the LiteLLM PyPI Supply Chain Attack
What Actually Happens Under the Hood When You Prompt an AI Coding Agent
Why we built our own sandboxing system
Case Study: AI-Driven Planning System for Case Workers
Products & Services
Open-source, container-free, deny-by-default sandbox for AI coding agents and other commands. It enforces filesystem, network and syscall controls on Linux and macOS, supports learning mode and built-in profiles, and works with the user's normal development environment.
SOCKS5/DNS and transparent network proxy supporting Greywall. It provides a live dashboard, network activity visibility, domain rules, dynamic allow/deny authorization and conversation tracking for AI-provider traffic.
Self-hostable meeting-audio application that uses local models to transcribe and analyze recordings. It supports local Ollama models, local storage, browser recording and file inputs, authentication and optional Zulip integration.
A single Go-binary LLM proxy positioned as a focused alternative to LiteLLM, usable with OpenAI-compatible SDKs and container deployment.
Market Position
Greyhaven positions itself as a sovereign-AI engineering partner rather than a hosted model vendor: it embeds engineers with customers and deploys systems inside their environments. For agent security, Greywall/Greyproxy competes with lightweight OS-level sandboxes such as Anthropic's sandbox-runtime and Fence, while differentiating through a real-time dashboard, dynamic network authorization, learning mode and transparent network capture. Its own comparison identifies cloud microVM approaches such as E2B, Vercel v0 sandbox, Together, RunLoop and Daytona, and gVisor-based offerings such as Modal and Google's Agent Sandbox as alternatives for different threat models.
Leadership
Founders
Max McCrea
CEO and founder; expert data scientist, former professional poker player, full-stack engineer and AI researcher. He is also a co-founder of Monadical, which he has built since 2016, and an alumnus of the Recurse Center.
Executive Team
Max McCrea
CEO / Founder
Expert data scientist, former professional poker player, full-stack engineer and AI researcher; co-founder of Monadical.
Mathieu Virbel
CTO
Full-stack developer focused on AI, DevOps, embedded systems and software/music design.
Founding Story
Greyhaven grew out of Monadical's decade-plus of building secure, self-hosted systems. The team wanted AI systems that preserve customer control over data and workflows, and created Greywall after finding that container-based and conventional proxy approaches were too brittle, disruptive to developers, or insufficiently observable. Their initial vision was sovereign AI: private assistants, workflow automation and knowledge systems deployed inside the customer's own environment.
Business Model
Revenue Model
Enterprise and organization engagements for embedded AI engineering, discovery, prototyping, deployment and extension of private/self-hosted systems; Greyhaven's terms refer to commercial services and customer-specific written agreements. Open-source tools such as Greywall are distributed for free, while the company says it is building a managed governance layer for teams running Greywall.
Target Markets
- SME teams whose decisions rely on institutional knowledge
- Public-sector agencies and regulated environments where data must stay within the perimeter
- Research and technical organizations with repeated analysis workflows
- Infrastructure and service providers needing private internal tooling
- Telecommunications and enterprise software organizations
- Organizations with privacy-sensitive meeting, knowledge or operational data
- Private AI assistants and workflow automation
- AI coding-agent sandboxing and observability
- Meeting transcription and analysis using local models
- Knowledge systems and repeated analysis workflows
- Enterprise software modernization and legacy COBOL/data-workflow transformation
- Case-worker planning and social-mobility programs
- A US-based nonprofit working with public-sector agencies
- A major telecommunications provider
- Organizations across North America, Europe and Latin America served through Monadical's secure self-hosted systems work