NOFire AI has released Brig as an Apache 2.0 open-source project, providing a microVM sandbox designed to run AI coding agents on Mac and Linux systems with hardware-enforced security boundaries.
The project targets a specific security gap that emerges when developers grant AI agents elevated permissions. These agents can install packages, execute arbitrary commands, access networks, and interact with developer credentials. NOFire AI points to the Axios supply chain attack as an example of infiltration vectors that become possible under broad system access. The company argues that as developers increasingly use auto-approval features to speed up their workflow, the entire operating system becomes a single point of failure.
Brig addresses this by isolating AI agents within an ephemeral microVM, where security is enforced at the hardware level. Memory and process boundaries are physically enforced by the underlying CPU architecture rather than relying on software isolation alone.
Leadership and Development Experience
The project’s co-founder and chief scientist, Anastasios Nanos, cited over a decade of microVM development experience. “The VM and microVM technologies I’ve worked on since 2007 are now more relevant than ever,” Nanos said. “The rise of AI agents and agentic workloads requires the security and trust boundary to move lower without paying a significant performance tax.” Nanos previously led development of urunc, a CNCF Cloud Native Compute Foundation Sandbox project focused on unikernel technologies.
Codebase and Security Advantages
The company positions the small codebase as a security advantage. The microVM core contains fewer than 20,000 lines of code, which NOFire AI says makes it auditable for security researchers without the complexity of larger sandbox technologies.
Brig supports curated profiles for Claude Code, Codex, Cursor, Gemini, Grok, and opencode. Users can also provide their own configurations using OCI images such as Ubuntu. The project runs on macOS with Apple Silicon and Linux on both x86-64 and ARM architectures.
Installation requires a single command: curl -fsSL https://brig.sh/install | sh.
Company Vision
NOFire AI‘s broader mission involves building what the company describes as a “Context and Control Model for Production”—a live, time-versioned system for monitoring production behavior with governance over agentic actions. According to co-founder and CTO Moustafellos Panagiotis, the company invested heavily in isolation technology that now extends beyond their internal use case.
Documentation, security details, and the source repository are available at brig.sh.