Polaxis

Polaxis is a runtime control layer for AI agents, functioning as an AI agent firewall and control plane. The product intercepts every tool call made by AI agents before execution, applying a 7-layer detection pipeline to evaluate and enforce security policies. The 7-layer pipeline consists of: - Regex scan: detects injection attempts, PII, and secrets across all calls - Risk scorer: evaluates 15 signals including entropy, delimiters, and payload size - LLM gate: applies an LLM-based evaluation to approximately 11% of calls that require deeper analysis - Behavioral baseline: per-agent anomaly detection using Welford online statistics - Session graph: multi-turn kill-chain detection across agent sessions - Threat intel: aggregates attack history and per-agent threat levels - Policy engine: JSON-based rules for blocking, allowing, or escalating tool calls Key capabilities include human-in-the-loop (HITL) approvals via email or Slack, budget and spend controls per agent, immutable audit trails, and on-demand compliance reporting for SOC 2, GDPR, HIPAA, EU AI Act, and OWASP. PII is automatically masked during human review workflows. Polaxis is distributed as a Python SDK and supports major agent frameworks including LangChain, LangGraph, CrewAI, OpenAI Agents SDK, PydanticAI, AutoGen, and the MCP protocol. It also offers an MCP proxy for zero-code setup. Target markets include fintech, healthcare, SaaS companies, and enterprise IT teams deploying AI agents. Pricing follows a tiered SaaS model with a free starter tier, Pro at $149/month, Scale at $499/month, and custom Enterprise plans with self-hosted (VPC) deployment options.