Pebblo is an AI agent governance and security platform developed by Daxa.ai. It is designed to control and secure autonomous AI agents operating within enterprise environments, particularly those using the Model Context Protocol (MCP) to access data sources and perform actions. Pebblo provides two primary modules: **Safe Infer:** An inline control point for IDE-to-model traffic. It inspects code snippets, prompts, and completions in real time, blocking sensitive content, logging events, and redacting information based on policy. It enforces an allow-list of approved models and validates responses before execution. **Safe MCP (Safe Agent):** An integrated module that includes an MCP Gateway and a Trust Zone where MCP servers are hosted. It handles vetting, sandboxing, runtime permissions, and supply-chain containment. It performs permission checks, role and project context validation, and payload sanitization for MCP-connected tools. Additional capabilities include: - Detection and stripping of malicious command or prompt injections before they reach agents or tools - Sanitization of payloads and tool outputs - A centralized Policy Plane for defining what data can be shared, which models are allowed, and which agent actions are permitted - Data classification at ingestion, tagging content for compliance frameworks such as GDPR and HIPAA - User access control at retrieval time, evaluating queries against document intent and user context - Full audit visibility across agent interactions Pebblo addresses risks including sensitive data exposure, unsafe autonomous actions, and over-permissioned access. It is designed to be transparent to developers, requiring no agent code changes.