Skip to content

NVIDIABackendAdapter

Defined in: backend/src/providers/nvidia.ts:67

Backend adapter for NVIDIA NIM API.

NVIDIA NIM (NVIDIA Inference Microservices) provides optimized inference with OpenAI-compatible API. Supports various models including Llama, Mistral, and other popular open-source models.

import { NVIDIABackendAdapter } from 'ai.matey';
const adapter = new NVIDIABackendAdapter({
apiKey: process.env.NVIDIA_API_KEY,
});
const adapter = new NVIDIABackendAdapter({
apiKey: process.env.NVIDIA_API_KEY,
});
const response = await adapter.execute({
messages: [{ role: 'user', content: 'Hello!' }],
parameters: {
model: 'meta/llama-3.1-405b-instruct',
},
});
const adapter = new NVIDIABackendAdapter({
apiKey: process.env.NVIDIA_API_KEY,
baseURL: 'https://your-nim-endpoint.nvidia.com/v1',
});
const stream = adapter.executeStream({
messages: [{ role: 'user', content: 'Tell me a story' }],
parameters: {
model: 'nvidia/llama-3.1-nemotron-70b-instruct',
},
});
for await (const chunk of stream) {
if (chunk.type === 'content') {
console.log(chunk.delta);
}
}

new NVIDIABackendAdapter(config): NVIDIABackendAdapter

Defined in: backend/src/providers/nvidia.ts:71

BackendAdapterConfig

NVIDIABackendAdapter

OpenAIBackendAdapter.constructor

protected readonly baseURL: string

Defined in: backend/src/providers/openai.ts:162

OpenAIBackendAdapter.baseURL


protected readonly config: BackendAdapterConfig

Defined in: backend/src/providers/openai.ts:161

OpenAIBackendAdapter.config


readonly metadata: AdapterMetadata

Defined in: backend/src/providers/openai.ts:160

Adapter metadata for identification and capabilities.

BackendAdapter.metadata

OpenAIBackendAdapter.metadata

estimateCost(request): Promise<number | null>

Defined in: backend/src/providers/nvidia.ts:127

Estimate cost for NVIDIA NIM.

IRChatRequest

Promise<number | null>

BackendAdapter.estimateCost

OpenAIBackendAdapter.estimateCost


execute(request, signal?): Promise<IRChatResponse>

Defined in: backend/src/providers/openai.ts:221

Execute non-streaming chat completion request.

IRChatRequest

AbortSignal

Promise<IRChatResponse>

BackendAdapter.execute

OpenAIBackendAdapter.execute


executeStream(request, signal?): IRChatStream

Defined in: backend/src/providers/openai.ts:260

Execute streaming chat completion request.

IRChatRequest

AbortSignal

IRChatStream

BackendAdapter.executeStream

OpenAIBackendAdapter.executeStream


fromIR(request): OpenAIRequest

Defined in: backend/src/providers/openai.ts:605

Convert IR request to OpenAI format.

Public method for testing and debugging - see what will be sent to OpenAI.

IRChatRequest

OpenAIRequest

BackendAdapter.fromIR

OpenAIBackendAdapter.fromIR


healthCheck(): Promise<boolean>

Defined in: backend/src/providers/nvidia.ts:107

Health check for NVIDIA NIM API.

Promise<boolean>

BackendAdapter.healthCheck

OpenAIBackendAdapter.healthCheck


invalidateModelCache(): OpenAIBackendAdapter

Defined in: backend/src/providers/openai.ts:591

Invalidate the cached model list.

Forces the next listModels() call to fetch fresh data from the API (unless static models are configured).

OpenAIBackendAdapter

This adapter for method chaining

OpenAIBackendAdapter.invalidateModelCache


listModels(options?): Promise<ListModelsResult>

Defined in: backend/src/providers/openai.ts:501

List available models from OpenAI.

This method supports three sources:

  1. Static config (config.models) - highest priority
  2. Cache (if enabled and not expired)
  3. Remote API (/v1/models endpoint)

ListModelsOptions

Promise<ListModelsResult>

BackendAdapter.listModels

OpenAIBackendAdapter.listModels


toIR(response, originalRequest, latencyMs): IRChatResponse

Defined in: backend/src/providers/openai.ts:652

Convert OpenAI response to IR format.

Public method for testing and debugging - parse OpenAI responses manually.

OpenAIResponse

IRChatRequest

number

IRChatResponse

BackendAdapter.toIR

OpenAIBackendAdapter.toIR