generated: '2026-06-20'
method: searched
source: https://build.nvidia.com; PyPI registry; NVIDIA developer docs
notes: >-
  NVIDIA NIM exposes an OpenAI-compatible REST surface, so the primary client is the
  standard OpenAI SDK pointed at https://integrate.api.nvidia.com/v1. NVIDIA additionally
  co-maintains a first-party LangChain integration and ships dedicated Python clients for
  the Riva speech gRPC surface, Triton, and the NeMo toolkit. The container CLI (ngc) is
  catalogued under cli/.
packages:
  - language: python
    registry: pypi
    name: langchain-nvidia-ai-endpoints
    url: https://pypi.org/project/langchain-nvidia-ai-endpoints/
    install: pip install -U langchain-nvidia-ai-endpoints
    official: true
    description: >-
      First-party LangChain integration (ChatNVIDIA, NVIDIAEmbeddings, NVIDIARerank) for
      the hosted NVIDIA API Catalog and self-hosted NIM microservices; connects to
      integrate.api.nvidia.com by default or a local NIM via base_url.
  - language: python
    registry: pypi
    name: nvidia-riva-client
    url: https://pypi.org/project/nvidia-riva-client/
    install: pip install nvidia-riva-client
    official: true
    description: >-
      Python client for NVIDIA Riva speech services (ASR, TTS, NMT) served by the Speech NIMs
      over gRPC. Source at github.com/nvidia-riva/python-clients.
  - language: python
    registry: pypi
    name: tritonclient
    url: https://pypi.org/project/tritonclient/
    install: pip install tritonclient[all]
    official: true
    description: >-
      NVIDIA Triton Inference Server client (HTTP/gRPC); Triton is one of the inference
      backends NIM microservices ship.
  - language: python
    registry: pypi
    name: nemo_toolkit
    url: https://pypi.org/project/nemo_toolkit/
    install: pip install nemo_toolkit
    official: true
    description: >-
      NVIDIA NeMo framework for building, customizing, and deploying the generative AI
      models served by NIM.
  - language: python
    registry: pypi
    name: openai
    url: https://pypi.org/project/openai/
    install: pip install openai
    official: false
    description: >-
      Standard OpenAI Python SDK, compatible with NIM when base_url is set to
      https://integrate.api.nvidia.com/v1. Documented by NVIDIA as the recommended client
      but published by OpenAI, not NVIDIA.