Cencurity is a self-hosted security gateway that sits between your applications and LLM providers. It proxies AI agent traffic, inspects both inbound tool calls and outbound model responses, and enforces security policies in real time. It can mask sensitive data like PII before it reaches upstream providers, block dangerous code patterns such as os.system or eval, and maintain full audit logs of every policy match. Built with a Go backend and React dashboard, it deploys via Docker Compose and supports OpenAI, Anthropic, and Gemini-compatible endpoints.

• Centralized Security Dashboard: Monitor every AI agent call in one place with real-time visibility into requests, responses, latency, policy hits, redactions, and blocks.
• Real-Time Protection: Automatically detect and block sensitive information, such as secrets and personal data, before it reaches users or models.
• Real-Time Log Analysis: Trace every interaction end-to-end. Use search, filter, and correlation tools to connect agent calls with policy decisions and identify risks quickly.
• Policy-First Detection: Identify policy violations rapidly and prioritize the issues most critical to your security posture.
• Zero-Click Guardrails: Implement safety controls that do not hinder the speed of development or delivery.
• Audit-Ready Reporting: Generate clear logs and evidence for use in compliance, reviews, or investigations.
• LLM Proxy with Redaction: Proxy model traffic and automatically remove sensitive data to ensure it is never exposed.
• Webhook Notifications: Receive verified alerts for critical security events within tools such as Slack and Jira.
• Dry-Run Rollout Mode: Test the behavior of security policies before full enforcement to measure impact and adjust settings safely.
• Personal API Keys and Dashboards: Issue unique API keys, proxy URLs, and dashboards for individual users to protect credentials and isolate access.

• Full Data Sovereignty: Self-hosted via Docker means your LLM traffic and audit logs never leave your infrastructure.
• Open-Source Community Edition: Free to deploy and inspect, giving you full visibility into how your security policies are enforced.
• Detailed Audit Trail: Every policy match is logged with context, supporting compliance and incident response workflows.
• Centralized Security Monitoring: View all agent calls in a single dashboard with real-time tracking of requests, responses, latency, policy hits, redactions, and blocks to help teams spot issues quickly.
• Real-Time Protection Against Leaks: Detect and block sensitive data, such as secrets and personal information, automatically before it reaches users or models.
• Policy-First Detection: Focus on the most critical security issues by efficiently prioritizing policy violations.
• Dry-Run Policy Mode: Test how policies behave before enforcement to refine controls without disrupting existing workflows.
• Verified Alerts via Webhooks: Send notifications for key security events directly to tools like Slack and Jira for faster response times and better team visibility.

• No Managed or Cloud Offering: Self-hosting only, which means your team handles infrastructure, updates, and maintenance
• Integration Gaps: No clearly advertised native exports to external tools like SIEMs or observability stacks.

• Protecting sensitive data from LLM providers: When your agents process customer data, PII, or internal credentials, Cencurity masks that information at the gateway before it's sent upstream. Your LLM provider never sees the raw sensitive data, helping you meet privacy and compliance requirements.
• Audit logging for regulated industries: Teams in finance, healthcare, or government that deploy AI agents need a full record of what data flowed through their LLM interactions. Cencurity's audit logs capture every policy match with severity and context, giving compliance teams the documentation they need.
• Engineering teams building agentic workflows: When AI agents interact with production APIs and databases, a single prompt injection could trigger unintended actions. Cencurity provides fail-closed enforcement, blocking unsafe tool calls and giving your team visibility into what agents are actually doing.