Agenta is an open-source workspace for building and running AI agents and LLM applications, used by teams to take agents from prototype to production. It began as an LLMOps platform for prompt engineering, evaluation and observability, and has grown into a full agent workspace. The project is developed in the open on GitHub, with more than 4,800 stars, and can be self-hosted or used as a managed cloud.
Agenta gives teams a shared place to build agents that act as AI coworkers: agents can be created from templates or AGENTS.md-style definitions, connected to more than a thousand integrations and MCP tools, and reached where work happens, including Slack and Telegram. Agents run on schedules or triggers, ask for human approval before sensitive actions, and keep credentials in a secret vault they can use but not read. Underneath, the original LLMOps strengths remain: a prompt playground, versioning, automated evaluations and tracing so you can compare variants and debug failures instead of shipping blind.
Teams can start in the cloud within minutes and migrate to a self-hosted deployment later without changing their agent definitions, and every agent run is logged end to end, which simplifies debugging and auditing in production.
Freemium. Agenta’s core is open source under the MIT license and can be self-hosted for free with your own models or subscriptions. Agenta Cloud offers a managed version with paid plans for teams that prefer not to run the infrastructure themselves.
Agenta is for product and engineering teams building LLM applications and workplace agents who need evaluation, observability and safe, auditable automation rather than a pile of disconnected scripts.