Products

Narrow promises. Clear boundaries.

Virasai AI work is at different stages of maturity. We make that visible so visitors can tell what is open source, what is being validated, and what is still ahead.

Early / public repository

Sentinel

Deterministic, non-executing evidence and change monitoring for public npm MCP servers — built to surface the release where a package you already trusted becomes something else.

Sentinel inspects an exact public npm artifact without executing it, emits a schema-validated report of what that artifact declares, and compares reports to show what changed. The repository also includes Watch: a self-hosted monitor that polls packages you choose, runs Sentinel on new releases, and raises a reviewable change notice.

It does not decide whether a package is safe or malicious. It does not inspect private packages, local source trees, production credentials, MCP tool-call contents, or live traffic.

Join waiting list

MAGUS Core

A research-led governance architecture for people operating their own agent infrastructure.

The MAGUS research describes structural alignment drift in long-running agentic systems and proposes bounded governance mechanisms. Core is being shaped for operators who need governance closer to the execution boundary.

Relevant to teams operating long-lived agents with consequential tool access, builders exploring explicit permissions and evidence over time, and research/prototyping work around execution-boundary controls.

Coming soon

Cloud Proxy

A future product and research track for agent users who cannot run local enforcement.

Its design deliberately faces a different trust model because it would sit in the traffic path. We are not taking sign-ups or making availability promises for it yet.