Matrix OS provides an AI-native cloud computer where users and their coding agents work together in persistent, inspectable workspaces. The platform runs agents like Claude, Codex, and Cursor in parallel sessions with live state, repos, and previews, keeping work interruptible and human-controlled. It is designed for teams that need always-on agent operations without surrendering data ownership or review authority.
Funding
Funding not disclosed
Founders
Product
Problem
AI coding agents typically operate through ephemeral chat sessions that lack persistent project state, forcing users to manually coordinate context across IDEs, terminals, logs, and task trackers. This fragmentation makes long-running agent work fragile, difficult to supervise, and hard to trust, especially when agents need to work asynchronously or across team boundaries.
Solution
Matrix OS provides a private cloud computer where users and their AI agents work together in a persistent, inspectable environment. The platform provisions an isolated machine with repos, files, terminals, and app state that remain live even when a laptop sleeps or a browser tab closes. Users can run multiple agents—including Claude, Codex, Cursor, OpenCode, Pi, and Gemini CLI—in parallel sessions, track what each agent is reading, editing, testing, and previewing, and merge only the work that survives human review. A resident agent called Hermes handles the operating-system layer around coding agents, managing tool connections, scheduled workflows, notifications, approvals, and memory. The workspace is owner-controlled and exportable, with data stored in the user's own database and runtime rather than a shared index.
Target Audience
Primary users are software development teams, project-driven firms, and enterprise organizations that need persistent, permission-aware AI agent infrastructure for coding, deal operations, KYC intake, reporting, and document workflows.
Features
- Persistent cloud computer with live repos, shells, files, app state, previews, and logs that remain accessible across devices
- Parallel agent orchestration across Claude, Codex, Cursor, OpenCode, Pi, and Gemini CLI sessions without blocking the local machine
- Hermes resident agent for non-coding operational work: tool connections, scheduled workflows, notifications, approvals, and memory
- Shared agent workspaces for teams with organization membership, project ownership, agent identities, roles, handoffs, and review
- Integration with OneDrive and Google Drive as working memory via provider APIs, change feeds, stable file identity, permissions, indexing, and controlled write-back
- Trust layer with scoped permissions, action-bound approvals, evidence-rich audit records, prompt-injection boundaries, and revocation
- Workflow engine that executes trigger → gather context → agent work → policy check → human review → action → shared record loops
- AGPL-3.0-or-later open-source codebase with hosted service available