NVIDIA NIM Provider

Point an agent at NVIDIA NIM with one llm string and one API key.

Quick Start

Simple Usage

export NVIDIA_API_KEY=nvapi-...

import { Agent } from 'praisonai';

const agent = new Agent({
  instructions: 'Be brief.',
  llm: 'nvidia/nemotron-super',
});

await agent.chat('Hello in one sentence.');

With Per-Agent Credentials

import { Agent } from 'praisonai';

const agent = new Agent({
  instructions: 'Be brief.',
  llm: 'nvidia/nemotron-super',
  apiKey: process.env.MY_NVIDIA_KEY,
  baseURL: 'https://integrate.api.nvidia.com/v1',
});

Per-agent apiKey and baseURL take effect on nvidia-nim as of PraisonAI PR #5831. Before that, they were silently ignored for aliased providers such as nvidia.

Model Names

NIM model ids always carry their publisher; short aliases map to the catalog slug, and a bare name under the nvidia prefix gets nvidia/ reattached. nvidia/, nim/ and nvidia-nim/ all select the same provider.

You write	Sent on the wire
`nvidia/nemotron-super`	`nvidia/nemotron-3-super-120b-a12b`
`nvidia/nemotron-nano`	`nvidia/nemotron-nano-3-30b-a3b`
`nvidia/nemotron-ultra`	`nvidia/nemotron-3-ultra-550b-a55b`
`nvidia/nemotron-3-super-120b-a12b`	`nvidia/nemotron-3-super-120b-a12b`
`nvidia/meta/llama-3.3-70b-instruct`	`meta/llama-3.3-70b-instruct`
`nim/nemotron-super`	`nvidia/nemotron-3-super-120b-a12b`
`nvidia-nim/nvidia/nemotron-3-super-120b-a12b`	`nvidia/nemotron-3-super-120b-a12b`

Environment Variables

# Hosted (build.nvidia.com)
export NVIDIA_API_KEY=nvapi-...

# Optional overrides
export NVIDIA_NIM_API_KEY=nvapi-...                   # takes precedence over NVIDIA_API_KEY
export NVIDIA_NIM_API_BASE=http://localhost:8000/v1   # any OpenAI-compatible NIM endpoint

Key order: explicit apiKey → NVIDIA_NIM_API_KEY → NVIDIA_API_KEY. OPENAI_API_KEY is never used.
Base-URL order: explicit baseURL → NVIDIA_NIM_API_BASE → https://integrate.api.nvidia.com/v1.

If no key is set and the base URL is the hosted endpoint, calls throw AUTHENTICATION. Get a key at build.nvidia.com, or pass a baseURL for a local NIM.

The provider uses the optional peer dependency @ai-sdk/openai-compatible. Install it if your package manager does not pull it in: npm install @ai-sdk/openai-compatible.

Reasoning Mode (Nemotron 3)

Nemotron 3 on NIM switches thinking through chat_template_kwargs, and the SDK translates reasoningEffort for you.

import { Agent } from 'praisonai';

const agent = new Agent({
  instructions: 'Be brief.',
  llm: 'nvidia/nemotron-super',
  reasoningEffort: 'off', // or 'minimal' | 'low' | 'medium' | 'high'
});

`reasoningEffort`	Sent to NIM
`"off"`	`chat_template_kwargs: { enable_thinking: false }`
`"minimal"` or `"low"`	`chat_template_kwargs: { enable_thinking: true, low_effort: true }`
`"medium"` or `"high"`	`chat_template_kwargs: { enable_thinking: true }`
unset	nothing — server default (thinking on)

This mapping only applies to Nemotron 3 (and nemotron-nano-3) via the NIM route. Nemotron served by other providers, and non-Nemotron models on NIM, use the standard reasoning_effort path.

Local NIM Container

Point baseURL at your own NIM container; no key is needed.

import { Agent } from 'praisonai';

const agent = new Agent({
  instructions: 'Be brief.',
  llm: 'nvidia/nemotron-super',
  baseURL: 'http://localhost:8000/v1',
});

Any baseURL outside integrate.api.nvidia.com is treated as a local or self-hosted NIM, and the SDK sends a placeholder key if none is set. Set NVIDIA_NIM_API_BASE to the same URL to make it the default.

Tools and Streaming

Tools
Streaming

import { Agent } from 'praisonai';

const getWeather = (city: string) => `Weather in ${city}: 20C, sunny`;

const agent = new Agent({
  instructions: 'Report the weather using the tool.',
  llm: 'nvidia/nemotron-super',
  reasoningEffort: 'off',
  tools: [getWeather],
});

await agent.start('Weather in Paris?');

import { Agent } from 'praisonai';

const agent = new Agent({
  instructions: 'You tell stories.',
  llm: 'nvidia/nemotron-super',
  reasoningEffort: 'off',
});

for await (const token of agent.stream('Tell me a short story about a robot')) {
  process.stdout.write(token);
}

reasoningEffort: 'off' keeps thinking tokens out of the output.

NVIDIA NIM CLI Usage

Check your NIM setup from the terminal

Providers

All supported providers

Model Routing

How llm strings resolve to providers

​Quick Start

​Model Names

​Environment Variables

​Reasoning Mode (Nemotron 3)

​Local NIM Container

​Tools and Streaming

​Related

NVIDIA NIM CLI Usage

Providers

Model Routing

Quick Start

Model Names

Environment Variables

Reasoning Mode (Nemotron 3)

Local NIM Container

Tools and Streaming

Related