AI - Artificial Intelligence

Resource for integration with Artificial Intelligence services, enabling the use of AI clients for text generation and embeddings, as well as vector stores for semantic storage and search.

Configuration example:

"ai": {
    "client": {
        "default": {
            "url": "http://127.0.0.1:11434/v1",
            "key": "ollama"
        }
    },
    "vector": {
        "default": {
            "engine": "file",
            "path": "vector_store.json"
        }
        "pg": {
            "engine": "pg",
            "host": "localhost",
            "port": 5432,
            "name": "netuno_vector",
            "username": "netuno_user",
            "password": "your_password"
        }
    }
}

The ContextRetrievalChunker allows splitting text into fragments for indexing and contextual retrieval.

// Example using the AI client with the default provider
const client = _ai.client();
const response = client.chat('Hello, how can I help?');
_log.info('Response: ' + response);

// Example using the vector store
const vector = _ai.vector();

if (!vector.collectionExists('netuno')) {
    vector.createCollection('netuno', 768);
}

---

client

---

_ai.client() : Client

Description

Gets the AI client using the default provider (default) defined in configuration.

How To Use

const client = _ai.client();
const response = client.chat('Hello!');
_log.info('Response: ' + response);

Return

( Client )

AI client instance configured with the default provider.

---

_ai.client(provider: string) : Client

Description

Gets the AI client for a specific provider.

How To Use

// Use a specific provider
const client = _ai.client('anthropic');
const response = client.chat('What is the capital of Portugal?');
_log.info('Response: ' + response);

Attributes

NAME	TYPE	DESCRIPTION
provider	string	Name of the AI provider configured under ai.client in the application configuration file.

Return

( Client )

AI client instance configured for the specified provider.

---

contextRetrievalChunker

---

_ai.contextRetrievalChunker() : ContextRetrievalChunker

Description

Creates a new ContextRetrievalChunker instance, a utility for splitting text into chunks suitable for indexing in vector stores and contextual retrieval in RAG (Retrieval-Augmented Generation) pipelines.

The chunker should be used during the document ingestion phase: the text is split into chunks, each chunk is converted into an embedding with client.embeddings and stored in the vector store with vector.add.

How To Use

// Ingestion phase: split document into chunks and index in the vector store
const chunker = _ai.contextRetrievalChunker();
const client = _ai.client();
const vector = _ai.vector('default');

if (!vector.collectionExists('netuno')) {
    vector.createCollection('netuno', 768);
}

const document = 'Long document text that will be split into chunks...';
const chunks = chunker.chunk(document);

for (let i = 0; i < chunks.size(); i++) {
    const chunk = chunks.get(i);
    const options = _val.init().set('encoding_format', 'float');
    const embeddingResponse = client.embeddings('embeddinggemma:latest', chunk, options);
    const embedding = embeddingResponse.get('data').get(0).get('embedding');
    const metadata = _val.init().set('source', 'document').set('index', i);
    vector.add('netuno', 'chunk-' + i, embedding, chunk, metadata);
}
_log.info('Indexed ' + chunks.size() + ' chunks.');

Return

( ContextRetrievalChunker )

New ContextRetrievalChunker instance for text chunking in RAG pipelines.

---

init

---

_ai.init() : AI

Return

( AI )

---

vector

---

_ai.vector() : org.netuno.tritao.ai.vector.VectorStore

Description

Gets the vector store using the default provider (default).

How To Use

const vector = _ai.vector();
const client = _ai.client();

// Create the collection if it does not exist yet (768 is the embedding model dimension)
if (!vector.collectionExists('netuno')) {
    vector.createCollection('netuno', 768);
}

// Generate the question embedding and search for the most relevant documents
const options = _val.init().set('encoding_format', 'float');
const embeddingResponse = client.embeddings('embeddinggemma:latest', 'What is Netuno?', options);
const questionEmbedding = embeddingResponse.get('data').get(0).get('embedding');

const docs = vector.search('netuno', questionEmbedding, 10);
for (const doc of docs) {
    _log.info('Score: ' + doc.get('score') + ' | ' + doc.get('content'));
}

Return

( org.netuno.tritao.ai.vector.VectorStore )

Vector store instance configured with the default provider.

---

_ai.vector(provider: string) : org.netuno.tritao.ai.vector.VectorStore

Description

Gets the vector store for a specific provider.

How To Use

const vector = _ai.vector('default');
const client = _ai.client();

// Create the collection if it does not exist yet (768 is the embedding model dimension)
if (!vector.collectionExists('netuno')) {
    vector.createCollection('netuno', 768);
}

// Register an MCP tool that uses the vector store for RAG
const schema = _val.init()
    .set('type', 'object')
    .set('properties', _val.init()
        .set('input', _val.init().set('type', 'string'))
        .set('topk', _val.init().set('type', 'number'))
    )
    .set('required', ['input']);

_mcp.registerTool(
    'retrieve-documentation',
    'Retrieves the most relevant documentation segments by computing cosine similarity between the query embedding and stored vectors.',
    schema,
    function(params) {
        const options = _val.init().set('encoding_format', 'float');
        const question = params.getString('input');
        const embeddingResponse = client.embeddings('embeddinggemma:latest', question, options);
        const questionEmbedding = embeddingResponse.get('data').get(0).get('embedding');
        const topk = params.getInt('topk', 40);
        const docs = vector.search('netuno', questionEmbedding, topk);
        return _val.init().set('success', true).set('message',
            _val.init().set('retrieved-documents', docs)
        );
    }
);

Attributes

NAME	TYPE	DESCRIPTION
provider	string	Name of the vector store provider configured under ai.vector in the application configuration file.

Return

( org.netuno.tritao.ai.vector.VectorStore )

Vector store instance for the specified provider. Uses FileVectorStore for the file engine and PostgreVectorStore for the pg, postgres, or postgresql engines.

---

vectorConfig

---

_ai.vectorConfig(name: string) : Values

Attributes

NAME	TYPE	DESCRIPTION
name	string

Return

( Values )

---