Vaikora › Blog › Product Updates
Eight milliseconds to enforce or allow: how Vaikora makes an AI agent policy decision before execution
At P50, Vaikora makes an enforcement decision in 8 milliseconds. That number matters because every AI agent action, every tool call, every data read or write, sits behind that decision point. Nothing executes until the policy engine has returned an answer.
This is the core idea behind AI runtime control: not reviewing what happened, but enforcing what is allowed before anything runs.
Why the execution boundary is the right place to enforce
An AI agent operates by calling tools. It reads files, queries databases, sends API requests, writes to external systems. Each of those is a discrete action with a discrete effect. If a policy violation gets caught after the action executes, the effect already happened. A file was read. A message was sent. A record was changed.
Vaikora sits inline between the agent and whatever it is calling. When the agent issues an action, Vaikora intercepts it, evaluates it against the configured policy, and returns one of four decisions: allow, block, modify, or require-approval. The action does not proceed until that decision comes back.
At P99, the decision takes 45 milliseconds. The gateway handles 10,000 or more actions per second. So the enforcement layer is not a bottleneck in practice, it is a gate.
What the policy engine actually checks
Vaikora ships six content inspection modules. When an action arrives, the gateway passes it through whichever modules apply: checking for sensitive data patterns, evaluating the target tool and scope, matching against configured allow or deny rules, and running any compliance preset logic.
The compliance presets cover HIPAA, PCI-DSS, GDPR, SOC 2, and ISO 27001. A team deploying an agent that handles payment card data can load the PCI-DSS preset, and any action that would expose or transmit cardholder data hits a block decision without custom rule authoring.
Identical input always produces an identical decision. There is no probabilistic element. The same action against the same policy returns the same result every time. That determinism is what makes the audit record meaningful.
The audit chain and what it proves
Every decision Vaikora makes is written to a SHA-256 hash-chained, tamper-evident audit log. Each entry links to the previous one cryptographically, so any modification to a past record breaks the chain.
This matters for two reasons. First, in an incident investigation you can reconstruct exactly what the agent was permitted or denied at each step, and you can trust that the log has not been altered. Second, in a compliance audit you can demonstrate that a specific policy was in force at a specific time and that every agent action was evaluated against it.
The log is not an optional reporting feature. It is produced for every decision, including allows.
How Vaikora fits into an existing stack
Vaikora does not replace your existing identity and access controls. It adds a runtime gate on top of Azure RBAC, AWS IAM, and Snowflake roles. Those controls set the permissions ceiling. Vaikora enforces policy at the action level within that ceiling, with the context the IAM layer does not have: what the agent is doing, with what data, against what tool, inside what session.
For integrations, Vaikora pushes AI agent threats into SentinelOne as threat intelligence indicators, severity-mapped with Critical at 100, High at 75, Medium at 50, and Low at 25. High-severity risk promotes into Azure Security Center and Microsoft Defender for Cloud as native incidents, with no custom engineering.
Vaikora also ships as an open-source, MIT-licensed MCP-compatible proxy called Vaikora Guard MCP, published on PyPI as `vaikora-guard-mcp`. Teams that want to self-host the enforcement layer can do that. The commercial control plane, hosted by Data443, adds centralized policy management, the audit dashboard, and the compliance preset library.
It supports 12 LLM providers with no SDK migration. You point it at the existing model API and enforcement applies immediately.
What to check next
If you are evaluating AI runtime control for your environment, start with one agent and one tool category. Map what that agent is permitted to call, what data it touches, and what a policy violation would look like. Then check whether your current setup would catch that violation before execution or only after.
The Vaikora Guard MCP proxy is on PyPI. You can run it locally against a test agent and inspect the enforcement decisions and audit log output before deploying anything to production.
See Vaikora enforce policy on your AI
Open-core AI runtime control. Self-host the MIT gateway free, or run the hosted Control Plane.
Get a demo Self-host the gateway
Vaikora