NVIDIA NeMo Fabric

NeMo Fabric gives applications one configurable, observable way to run applications across multiple agent harnesses. It standardizes configuration, lifecycle management, and results without requiring a separate integration for every harness.

NeMo Fabric lets you change harnesses without rebuilding each integration, isolate conflicting runtime dependencies, and manage harness configuration, execution, and observability consistently. Every run returns normalized results, artifacts, and telemetry for downstream systems to consume.

It provides:

• a versioned, typed configuration contract;
• ordinary Python composition for experiment variants;
• adapter integrations for harness-specific launch and control;
• a Python SDK backed by the Rust core;
• normalized run results, artifact manifests, and telemetry references.

Install

Install the core runtime and Python SDK:

pip install nemo-fabric

To use a supported agent harness, install its adapter extra:

pip install "nemo-fabric[claude]"
pip install "nemo-fabric[codex]"
pip install "nemo-fabric[deepagents]"
pip install "nemo-fabric[hermes]"

NeMo Fabric supports running an agent harness in a different virtual environment than the one used to run NeMo Fabric itself. This is useful for running agents that have conflicting dependencies with NeMo Fabric or other agents.

The adapter must be installed into the virtual environment that the harness is installed in. For this reason adapters intentionally have minimal dependencies.

Integrations

Harbor Integration

pip install "nemo-fabric[harbor]"

Relay Integration

pip install "nemo-fabric[relay]"

This installs a version of NeMo Relay Python library known to be compatible with the installed version of NeMo Fabric.

Some adapters, such as Claude and Codex, require the nemo-relay CLI tool instead of the NeMo Relay Python library. Refer to the NeMo Relay CLI install guide for instructions on installing the CLI tool.

Python Versions

NeMo Fabric supports Python versions 3.11-3.14, however some of the integrations and adapters may have additional requirements. Specifically Hermes Agent doesn't support Python 3.14 yet, and the Harbor integration requires Python 3.12 or later.

Core Concepts

• Typed configuration: Construct a complete FabricConfig in Python and use ordinary functions to create experiment variants.
• Adapters: Select harness-specific integrations with harness.adapter_id.
• Artifacts: Receive normalized output, logs, patches, and telemetry references through an ArtifactManifest.

The experimental nemo-fabric CLI is distributed separately from the Python package. It selects complete typed configs from built-in presets and maintained examples. Applications that need a stable integration surface should use the Python SDK.

Learn More

Refer to the NVIDIA NeMo Fabric documentation for installation, configuration, and usage guidance. Source code is available in the NVIDIA NeMo Fabric repository.