neuron-rag

Neuron AI RAG

This skill helps you implement Retrieval-Augmented Generation (RAG) in Neuron AI. RAG extends Agent — it inherits everything an agent can do (chat, stream, structured output, tools, persistence, thread identity) and replaces the entry chain with retrieval:

AgentStartEvent → PreProcessNode → RetrievalNode → PostProcessNode → InstructionsNode → inference

1. Pre-process the user question (query rewriting/expansion)
2. Retrieve relevant documents from the vector store
3. Post-process (re-rank, filter by score)
4. Send the documents with the question, as context of the turn, and run the normal inference

The documents are not added to the instructions and not stored in the chat history. They reach the model in the question's own message, after the question, inside <EXTRA-CONTEXT> tags:

system   <your instructions>
user     <the question>
         <EXTRA-CONTEXT> ...retrieved documents... </EXTRA-CONTEXT>

Write instructions that fit this position, such as "answer from the context that follows the question". When nothing is retrieved, the question is sent alone. Because the instructions no longer change with each question, they and the conversation can be cached: see "Turn context" in the neuron-agent skill.

Core Components

1. Embeddings provider — converts text to vectors (EmbeddingsProviderInterface)
2. Vector store — stores and searches document embeddings (VectorStoreInterface)
3. Retrieval strategy — how a query becomes a document list (RetrievalInterface)
4. Pre/post processors — query transformation and result re-ranking/filtering

use NeuronAI\Providers\AIProviderInterface;
use NeuronAI\Providers\Anthropic\Anthropic;
use NeuronAI\RAG\Embeddings\EmbeddingsProviderInterface;
use NeuronAI\RAG\Embeddings\OpenAIEmbeddingsProvider;
use NeuronAI\RAG\RAG;
use NeuronAI\RAG\VectorStore\PineconeVectorStore;
use NeuronAI\RAG\VectorStore\VectorStoreInterface;

class MyChatBot extends RAG
{
    protected function provider(): AIProviderInterface
    {
        return new Anthropic(
            key: $_ENV['ANTHROPIC_API_KEY'],
            model: 'ANTHROPIC_MODEL',
        );
    }

    protected function embeddings(): EmbeddingsProviderInterface
    {
        return new OpenAIEmbeddingsProvider(
            key: $_ENV['OPENAI_API_KEY'],
            model: 'OPENAI_EMBEDDING_MODEL',
        );
    }

    protected function vectorStore(): VectorStoreInterface
    {
        return new PineconeVectorStore(
            key: $_ENV['PINECONE_API_KEY'],
            indexUrl: $_ENV['PINECONE_INDEX_URL'],
        );
    }
}