NVIDIA Dynamo
Project record
Dynamo is NVIDIA's open-source framework for serving generative AI models across many GPUs and nodes. It sits above inference engines such as vLLM, SGLang, and TensorRT-LLM and adds disaggregated prefill and decode, KV-cache-aware request routing, KV cache offloading, and SLA-based autoscaling behind an OpenAI-compatible frontend. It is written in Rust and Python.14
- Repository: GitHub repository (external site: github.com)
- Documentation: Documentation (external site: docs.nvidia.com)
- License: LICENSE (Apache 2.0) (external site: github.com)
- Release notes: Release v1.5.0 (external site: github.com)
Availability and license
Overall availability
Source code is public on GitHub. The README documents prebuilt runtime containers on NVIDIA NGC, the ai-dynamo package on PyPI, and Kubernetes deployment through the Dynamo Platform.17
Availability is separate from permission: read the license before using or redistributing.
The LICENSE file applies Apache 2.0 to the codebase except for test data files under lib/llm/tests/data/deepseek-v3.2, which it says are derived from the DeepSeek-V3.2 model repository and are under the MIT License. Contributions must be signed off under the Developer Certificate of Origin. The license does not cover the models served with Dynamo.23
Component reuse rights
- weights
- Unknown — no complete fact-level rights review
- code
- Reviewed qualifying license recorded — check scope and conditions
- data
- Unknown — no complete fact-level rights review
- documentation
- Unknown — no complete fact-level rights review
No complete system-rights review is recorded for this release.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| Source codeIs the source code publicly readable? | Public | Developed in the public ai-dynamo/dynamo GitHub repository.1 |
| DocumentationIs user documentation published? | Public | NVIDIA publishes Kubernetes, local, and developer guides, recipes, and reference pages.4 |
| InstallationAre installation instructions or packages publicly available? | Public | The README documents prebuilt vLLM, SGLang, and TensorRT-LLM runtime containers, PyPI installation with backend extras, Kubernetes deployment, and building from source.17 |
| Supported platformsAre supported operating systems or hardware documented? | Public | For v1.5.0 the compatibility page lists NVIDIA Blackwell, Hopper, Ada Lovelace, and Ampere GPUs, Ubuntu 24.04 (Ubuntu 22.04 for wheels only), and x86_64 and ARM64 (ARM64 on Ubuntu 24.04 only). The documentation home page also mentions AMD GPUs and Intel XPUs.54 |
| Release statusAre versioned releases published? | Public | Versioned releases are published on GitHub. Dynamo 1.0 was announced in March 2026; v1.5.0, published on GitHub on September 21, 2026, was the current GA release when checked. Model-specific preview builds are also tagged.659 |
What it is useful for
The README positions Dynamo for serving models across multiple GPUs or nodes, for example when prefill and decode need to scale independently or requests should be routed by KV cache overlap; it notes that a single model on a single GPU is usually served by an inference engine alone.1
Run and use notes
- The README documents a single-node quick start that runs the Dynamo frontend and a vLLM or SGLang worker inside a prebuilt container with file-based discovery; etcd and NATS are not required for local or Kubernetes deployments.1
Organization context
U.S. eligibility
Eligible · basis: U.S.-governed project
NVIDIA announced Dynamo as its open-source inference software in March 2025 and publishes its documentation on docs.nvidia.com. The repository sits in the separate ai-dynamo GitHub organization, but its README and contribution guide carry NVIDIA Corporation & Affiliates copyright notices and no other governing entity is documented. NVIDIA's principal executive offices are in Santa Clara, California, per its Form 10-Q for the quarter ended July 26, 2026.841310
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.