Replicate Provider

Replicate runs open-source AI models in the cloud with a flexible adapter system.

Installation

npm install @aeye/replicate replicate

Configuration

import { ReplicateProvider } from '@aeye/replicate';
import { replicateTransformers } from '@aeye/models';

const replicate = new ReplicateProvider({
  apiKey: process.env.REPLICATE_API_KEY!,
  transformers: replicateTransformers,  // model adapters
  hooks: {
    chat: {
      beforeRequest: (request, input, ctx) => {},
      afterRequest: (request, input, response, ctx) => {},
      onError: (request, input, error, ctx) => {},
    },
  },
});

Model Adapters (Transformers)

Replicate models have diverse input/output schemas. Transformers adapt them to @aeye's unified interface:

import { replicateTransformers } from '@aeye/models';

// Pre-built adapters for popular models
const replicate = new ReplicateProvider({
  apiKey: '...',
  transformers: replicateTransformers,
});

The @aeye/models package includes adapters for hundreds of Replicate models.

Capabilities

Capability	Models
Chat	Llama 3, Mistral, etc.
Image Generation	SDXL, Flux, Stable Diffusion
Transcription	Whisper
Speech	Bark, etc.

Usage

const ai = AI.with()
  .providers({ replicate })
  .create();

// Chat with an open-source model
const response = await ai.chat.get(
  { messages },
  { metadata: { model: 'meta/meta-llama-3-70b-instruct' } }
);

Error Types

// Replicate uses the common provider error types
// Errors are caught and normalized in the adapter layer

Version Pinning

Replicate models can be pinned to specific versions for reproducibility. This is handled through the transformer configuration.