Retrievers¶

Sikka Agent's Retrievers module enables semantic search and intelligent information retrieval, allowing agents to find relevant information based on meaning rather than just keywords.

Overview¶

Effective retrieval is critical for grounding AI responses in accurate, relevant information. The Retrievers module provides:

Semantic search using vector embeddings
Contextual information retrieval from various sources
Hybrid search combining semantic and keyword approaches
Reranking capabilities for improved relevance
Multiple embedding model options (OpenAI, Sentence Transformers, Ollama, etc.)

These capabilities allow for implementing Retrieval Augmented Generation (RAG) patterns that enhance agent responses with specific information.

Embedding Models¶

Sikka Agent provides built-in support for various embedding models through the embedder module:

Base Embedder¶

The foundational class for all embedding models in Sikka Agent.

from sikkaagent.retrievers.embedder.base import Embedder

class CustomEmbedder(Embedder):
    dimensions: int = 768  # Set appropriate dimensions for your model

    def get_embedding(self, text: str) -> list[float]:
        # Implement embedding logic
        pass

    def get_embedding_and_usage(self, text: str) -> tuple[list[float], dict | None]:
        # Implement embedding with usage tracking
        pass

SentenceTransformerEmbedder¶

A wrapper for Sentence Transformers models that creates text embeddings.

Parameters¶

Parameter	Type	Description	Default
`id`	`str`	Name of pre-trained model	`"sentence-transformers/all-MiniLM-L6-v2"`
`dimensions`	`int`	Embedding dimensions	`384`
`sentence_transformer_client`	`SentenceTransformer`	Optional pre-initialized client	`None`

Returns¶

get_embedding(): Returns a vector embedding for text
get_embedding_and_usage(): Returns embeddings with usage information

Code Example¶

from sikkaagent.retrievers.embedder.sentence_transformer import SentenceTransformerEmbedder

# Create embeddings model with specific configuration
embedder = SentenceTransformerEmbedder(
    id="sentence-transformers/multi-qa-mpnet-base-dot-v1"
)

# Create embeddings for a query
query_embedding = embedder.get_embedding("How do neural networks work?")

OpenAIEmbedder¶

A wrapper for OpenAI's embedding models.

Parameters¶

Parameter	Type	Description	Default
`id`	`str`	Model ID	`"text-embedding-3-small"`
`dimensions`	`int`	Embedding dimensions	`1536`
`encoding_format`	`str`	Output format	`"float"`
`api_key`	`str`	OpenAI API key	`None`
`organization`	`str`	OpenAI organization	`None`
`base_url`	`str`	API base URL	`None`

Returns¶

get_embedding(): Returns a vector embedding for text
get_embedding_and_usage(): Returns embeddings with token usage information

Code Example¶

from sikkaagent.retrievers.embedder.openai import OpenAIEmbedder

# Create OpenAI embedder
embedder = OpenAIEmbedder(
    id="text-embedding-3-small",
    dimensions=1536,
    api_key="your-api-key"
)

# Generate embedding
embedding = embedder.get_embedding("What is machine learning?")

# Get embedding with usage statistics
embedding, usage = embedder.get_embedding_and_usage("What is machine learning?")
print(f"Tokens used: {usage}")

Other Embedders¶

Sikka Agent also provides:

OllamaEmbedder: For using Ollama's embedding models
FastEmbedEmbedder: For using the FastEmbed library

Rerankers¶

Sikka Agent provides reranking capabilities through the reranker module to improve retrieval quality:

Base Reranker¶

The foundational class for all rerankers in Sikka Agent.

from sikkaagent.retrievers.reranker.base import Reranker
from sikkaagent.document import Document

class CustomReranker(Reranker):
    def rerank(self, query: str, documents: list[Document]) -> list[Document]:
        # Implement reranking logic
        # Sort documents by relevance to the query
        return sorted_documents