Vaikora launches open-source proxy to secure AI agent tool calls

Vaikora has released Vaikora Guard MCP, an open-source, MIT-licensed, self-hostable proxy designed to provide inline security for AI agents. Unlike traditional security methods that focus on logging actions after they occur, this proxy sits between an AI agent and its tools—such as LLMs, APIs, or databases—to enforce policies before execution.

The system uses a deterministic policy engine to provide four possible outcomes for every tool call: allow, block, modify, or require-approval. This approach aims to prevent risks such as prompt injection, unauthorized data exfiltration, and privilege escalation by ensuring that no action proceeds until it is validated against a set policy.

Technical performance metrics indicate low latency, with a P50 enforcement latency of 8ms and a P99 of 45ms. The platform is capable of sustaining over 10,000 actions per second. Additionally, the system maintains a tamper-evident audit log using SHA-256 hash-chaining to ensure the integrity of all decision records.