Archestra is an enterprise platform designed to secure and govern AI agents operating within organizational environments. Its core product is a Kubernetes-native Model Context Protocol (MCP) orchestrator that enables enterprises to manage, monitor, and secure AI agents and their interactions with internal systems and data. The platform addresses a class of AI-specific vulnerabilities, notably prompt injection attacks that can lead to data exfiltration. Archestra describes this risk as the "Lethal Trifecta" — a combination of agent access to private data, processing of untrusted content, and external communication capability. To counter this, the platform provides deterministic guardrails that block unauthorized data flows and prevent agents from executing prompt injection instructions. Key platform capabilities include: - Private MCP Registry: A self-hosted registry for managing approved MCP servers with version control, granular access management, and compliance enforcement. - Security Guardrails: Deterministic controls to prevent data exfiltration, system corruption, and prompt injection exploitation by AI agents. - Centralized Governance: Secrets management, access control, audit trails, and team-based permissions across multi-user Kubernetes environments. - Cost Management: Per-team, per-agent, and organization-wide token usage monitoring, budget limits, and dynamic model switching to optimize LLM costs. - Observability: Integration with Prometheus, OpenTelemetry, and Grafana for metrics, distributed tracing, and performance monitoring. - User Interface: A ChatGPT-like chat interface supporting multiple LLM models (Claude, GPT-4, Gemini, open-source) with Slack, Microsoft Teams, and email integration. The platform is deployed via Helm charts on Kubernetes and supports Infrastructure as Code via a Terraform provider. It targets enterprise teams deploying AI agents at scale who require security, governance, and operational controls.