Aegisora
Open-source zero-trust runtime security and governance layer for autonomous AI agents
About Aegisora
Aegisora is an open-source runtime security and governance layer for autonomous AI agents. It treats agent tool calls, service access, workflow execution, and actions in external systems as runtime security boundaries, intercepting and evaluating each action before execution and returning an ALLOW, BLOCK, or ESCALATE decision. Explicit governance rules constrain agent behavior and permissions, limiting executable actions and reachable execution surfaces; the current repository integrations listed are OpenAI, Anthropic, and Gemini.
Best for
Best use cases
Evaluating actions before agents call tools
Controlling permissions when agents access services or external systems
Governing policies during agent workflow execution
Runtime security control across multiple model providers
Features
Runtime Interception
Evaluates actions before agent execution.
Policy Enforcement
Constrains agent behavior with explicit governance rules.
Permission Control
Limits executable actions and reachable execution surfaces.
Three-State Decision Model
Returns ALLOW, BLOCK, or ESCALATE based on security and governance requirements.
Pros
- Supports three runtime decisions: ALLOW, BLOCK, and ESCALATE
- Released under the MIT License
- Supports OpenAI, Anthropic, and Gemini integrations
- Emphasizes provider-agnostic security governance decisions at the runtime boundary