Quest: Open-source, safety-focused harness (github.com)

Quest has launched an open-source, safety-focused platform designed for building self-hosted AI agents. This innovative workbench allows users to connect their existing services—such as email and productivity tools—to various Large Language Models (LLMs) like Anthropic Claude and Google Gemini, while prioritizing data security. By using a user-friendly interface, Quest aims to empower non-engineers to leverage AI safely, enabling agents to interact with sensitive data without a risk of external exposure. Significantly, Quest emphasizes a strong security posture through several mechanisms: all conversations remain private, agents can only access predefined connected services, and a no-egress sandbox restricts external data movement. Each write operation requires explicit user approval, ensuring transparent interaction with connected applications. The platform's architecture not only includes shared workspaces and reusable skills but also implements a robust plugin system that allows customized integrations without compromising security. This structure opens up new possibilities for organizations looking to utilize AI efficiently while safeguarding sensitive information.