Exeora

Secure remote execution for AI agents on your own machines

Exeora connects MCP-compatible AI clients to projects running on a local computer, server, virtual machine, or build host. Its native CLI creates an outbound WebSocket connection to a Cloudflare-hosted gateway, avoiding inbound ports and keeping source files on their original machine.

Secure by Design

I designed Exeora around project-scoped access and OAuth credentials. Agents receive controlled file, search, editing, and command tools with path isolation, configurable policy modes, optional approvals, revocation, and metadata-only audit records.

The native execution layer is written in Rust, while the control plane and web applications use TypeScript, Bun, Cloudflare Workers, and generated protocol contracts. Scoped and short-lived authorization keeps remote execution useful without exposing a general-purpose machine credential.

Result

Exeora allows developers to use cloud-based AI clients with machines that cannot or should not expose inbound services. The source remains local while the agent receives only the capabilities explicitly granted to a project.