Platform / Policy Engine
Codify your agent boundaries
Stop relying on prompt engineering for security. Define strict, verifiable policies that govern exactly what your agents can and cannot do.
Available nowshipping with ActantOSComing soonon the product roadmap
Security as Code
Prompting an LLM to "never delete records" is not a security strategy. Modern LLMs can be tricked via prompt injection to ignore system instructions. The ActantOS Policy Engine moves security out of the model and into deterministic infrastructure.
Policies in ActantOS are defined as formal, deterministic rules — not prompt instructions. You define which endpoints can be called, what parameters are allowed, and how much an agent is permitted to spend. When an agent attempts an action, the proxy evaluates policy before the request ever reaches your backend. ActantOS uses AWS Cedar today; the engine is abstracted so additional policy backends can follow.
- Restrict access by agent identity (RBAC)
- Enforce hard spending limits per session
- Sanitize inputs and validate outputs
- Block specific tool parameters (e.g. destructive actions)
permit (
principal,
action,
resource
)
when {
resource.credential_access == false
};Core Capabilities
Declarative Policies
Available nowDefine agent boundaries in code using AWS Cedar. Five starter templates ship with v0.1.0 covering common enterprise patterns.
Zero-Trust Enforcement
Available nowEnforcement happens at the proxy layer, meaning agents can never bypass restrictions even if compromised.
Low-Latency Evaluation
Available nowPolicy evaluation is designed for minimal overhead so governance stays on the agent critical path.