Orca – Open-source runtime that turns any agent harness into a managed agent (github.com)

Orca has announced the release of the Orca Agent Engine, an open-source runtime that facilitates the management of AI agents. This self-hosted implementation allows developers to define and run agents, manage sessions, and stream event data while maintaining control over their infrastructure. Following closely on the heels of Anthropic's Managed Agents beta, Orca’s architecture consists of four key components: a REST and SSE API registry for metadata management, a runtime for executing agent loops, pluggable sandboxes for running session code, and an AI gateway to handle egress. The system also incorporates secure credential management, ensuring sensitive information remains protected. The significance of Orca lies in its potential to empower developers to create and manage their own AI agents in a customizable and controlled environment, streamlining the development process in the AI/ML landscape. With compatibility with existing Anthropic SDKs and the ability to operate within Docker and Kubernetes setups, Orca simplifies the complexities of agent management while supporting various runtime and storage solutions. The project is currently in developer preview with release 0.5.x, and its ongoing evolution could lead to substantial advancements in agent-based AI applications.