NVIDIA NeMo Fabric Overview
NVIDIA NeMo Fabric provides applications, evaluation systems, and rollout platforms a consistent interface for running agent harnesses and custom agents through a Python SDK backed by a Rust core.
- Configure runs: Select models and configure supported tools, skills, MCP servers, subagents, and runtime settings.
- Validate before execution: Check configuration compatibility before starting a runtime.
- Switch harnesses: Reuse the consumer integration while adapters handle harness-specific behavior.
- Compare outcomes: Collect normalized results and errors, along with available artifacts, lifecycle events, and telemetry references.
What You Can Do with NeMo Fabric
Use NeMo Fabric to integrate multiple supported harnesses into your product, compare agent configurations on the same tasks, inspect available execution evidence, or share a custom agent with other teams. The Key Features page describes these workflows and their measurement details.
How NeMo Fabric Works
Your application submits a task and configuration through the Python SDK. NeMo Fabric validates the settings, runs the selected harness through its adapter and environment provider, and returns a normalized outcome with available execution evidence. Your application owns scheduling, retries, scoring, and recovery. The Architecture Overview describes in more detail the component responsibilities and data flow.
Skip Ahead
Choose a starting point for your workflow: