Sondera

The company operates in the Large Language Model Operations (LLMOps) and AI security markets, addressing the risks associated with the increasing adoption of AI agents in corporate environments. These autonomous programs can access a user's computer, files, and online services to automate tasks, but they also introduce significant security vulnerabilities, such as data leakage, compliance failures, and prompt injection attacks. Sondera aims to provide organizations with the tools to manage these risks, enabling them to innovate with AI responsibly.

Sondera's core product is the Sondera Harness, an external framework that wraps around an organization's existing agent scaffolds (like LangGraph, Google ADK, Strands, or custom code) to monitor and control agent behavior. The Harness intercepts agent actions at six key stages of the execution loop: before the agent run starts, before and after a large language model (LLM) is called, and before and after a tool is executed. This allows companies to define and enforce security policies, written in the Cedar policy language, from a central point. These policies can prevent undesirable actions, such as data exfiltration or the execution of destructive commands, without being susceptible to bypass by prompt injection or other emergent behaviors from the AI agent itself.

The platform offers two deployment options: a cloud-based or self-hosted platform for managing policies across multiple agents with features like audit logs and team collaboration, and a local evaluation option for experimentation with fewer dependencies. By providing granular control and observability, Sondera helps organizations improve the reliability and security of their AI agents, steer them toward better performance, and ensure compliance.