AgentOven

AgentOven is an open-source enterprise agent control plane for managing AI agents. It combines a central agent registry, model routing, RAG pipelines, workflow orchestration, observability, and native A2A protocol support in one system. The site also describes it as the open-source Kubernetes for AI agents.

Its feature set includes registering, versioning, and discovering agents with rich metadata, along with first-class support for Google’s Agent-to-Agent protocol. Agent routing can span OpenAI, Anthropic, Azure, Bedrock, and local models, with fallback chains and cost caps. Other named capabilities include version-controlled prompt management with variable interpolation, A/B testing, and rollback, declarative DAG-based workflow orchestration with chaining, conditions, fan-out and fan-in, pause and resume, real-time token and API cost tracking with budget alerts and quotas, and OpenTelemetry-native distributed tracing, metrics, and structured logging. It also supports native MCP integration, zero-downtime blue/green deploys, auto-scaling, health checks, canary rollouts, built-in vector search, five retrieval strategies for RAG, and enterprise data connectors for Snowflake, Databricks, and S3.

Delivery is through a CLI and SDKs. The page shows agentoven-cli commands, an initialization flow that creates agentoven.toml and prompts/system.md, and SDKs for Python, TypeScript, and Rust. The architecture is described in layers, from SDKs and CLI through protocol and control-plane layers to infrastructure components including PostgreSQL, pgvector, Redis, OpenTelemetry Collector, and Prometheus/Grafana. Example code shows registration, baking an agent, and sending a task through A2A. The site also says the tool works with LangChain in the example and refers to workspace isolation through a kitchen model.

AgentOven is open source and MIT licensed. The page presents it as built for teams and organizations that need a single control plane for agent registry, routing, workflows, RAG, observability, cost tracking, and governance.