ZenML

A unified platform for ML pipelines and AI agent workflows that makes executions reproducible and agent failures replayable as regression tests, addressing versioning, lineage, caching, and evaluation of models and agents. It sells to developers, data teams, and operations teams in B2B organizations building and operating ML and AI systems. It is delivered as an open-source core that runs as a metadata layer on the customer's own infrastructure and VPC, with an optional managed control plane, and works with existing tools and clouds without requiring code changes between local and production.

Key features

• Replay agent traces as regression tests
• Reproducible ML pipeline orchestration
• Unified execution for ML and agent flows
• Artifact and checkpoint versioning
• Model registry and lineage tracking
• Infrastructure abstraction for any cloud
• Dockerization and GPU provisioning
• Smart caching and deduplication
• Reuse recorded tool calls on replay
• Centralized API key management
• RBAC and audit trails
• Execution trace visualization
• Run locally or on Kubernetes unchanged
• 60+ integrations across AI ecosystem