Aegisora 3.0 — Runtime Security for AI Agents, Now Open

AI agents are moving from generating answers to taking actions.

They call APIs, query databases, modify systems, and make decisions on behalf of users.

That creates a different security problem:

It’s not enough to know what the model generated.

We need to know what the agent was allowed to do, what decision was made, what actually executed, and what evidence remained afterward.

That’s what we’ve been building toward with Aegisora 3.0.

In this release, the runtime connects:

Request

→ Context

→ Policy

→ Decision

→ Execution

→ Evidence

The goal is simple: when an agent takes an unexpected action, you should have a traceable execution path instead of trying to reconstruct the incident from scattered logs.

Aegisora 3.0 is now open source:

https://github.com/aegisora-ai/aegisora/releases/tag/v3.0.0

We’d especially like feedback from people actually building agents.

Try it with a real workflow.

Break it.

Tell us what doesn’t make sense.

Open an issue, suggest an improvement, or contribute directly to the project.

We’re building this in public, and the best way for us to improve it is to see how other engineers use it in the real world.

GitHub stars and sharing the project with other agent builders also help us reach the people working on this problem.