Agenta 2.0 pivots from LLMOps evaluation to an open-source agent workspace treating agents as a compute primitive

Agenta·low signal

Announced July 22, Agenta 2.0 rebuilds the product as an open-source workspace for building and running agents, available as both cloud and self-hosted. Features include chat-based agent building with persistent workspaces, feedback-driven continuous improvement, team sharing, MCP server and skills support, multi-model routing across Claude/GPT/GLM, human-in-the-loop controls with detailed permissions, and full run logging and tracing. The stated thesis is that users describe jobs and supply context and tools rather than defining every workflow step. Single-source announcement with no independent coverage or adoption numbers yet.