Redis Vector Library (RedisVL)

RedisVL can be used to power AI agents by providing them with relevant information through several mechanisms:

• Retrieval-Augmented Generation (RAG): Use vector search and hybrid queries to retrieve context for LLMs.
• Memory: Implement persistent message history across different user sessions.
• Context Engineering: Curate the optimal context window by combining vector embeddings, complex metadata filtering, and reranking techniques.

Manage RedisVL via terminal or expose it to AI agents:

• CLI Management: Use the CLI to create, inspect, and delete indices directly from your terminal.
• Model Context Protocol (MCP):
  • Expose an existing Redis index to MCP clients.
  • Secure the MCP server by requiring JWT bearer tokens and gating read vs write permissions.

To reduce latency and costs in AI applications, RedisVL provides specialized tools:

• Semantic Caching: Use SemanticCache to cache LLM responses based on the semantic meaning of queries rather than exact string matches.
• Embeddings Caching: Use EmbeddingsCache to avoid redundant and expensive calls to embedding models by caching previously computed vectors.
• Semantic Routing: Use SemanticRouter to direct queries to the appropriate handler based on their semantic intent.

RedisVL enables search experiences that go beyond simple keyword matching:

• Semantic Search: Perform vector queries combined with complex metadata filtering.
• Hybrid Search: Combine traditional keyword search with vector search for improved accuracy.
• SQL Translation: Use the SQLQuery interface to interact with your data using familiar SQL syntax.

RedisVL provides several extensions to enhance LLM workflows:

• Semantic Caching: Use Cache LLM Responses to reduce costs and latency by caching semantic matches of LLM queries.
• Managed Caching: Integrate with LangCache as a managed semantic cache service.
• Message History: Use Manage LLM Message History to maintain persistent chat history with relevancy-based retrieval.
• Intent Routing: Use SemanticRouter to classify user intents and route queries to different handlers or models.

Manage how text is converted into vectors:

• Vectorizers: Create embeddings using various providers such as OpenAI, Cohere, HuggingFace, and others via the Create Embeddings with Vectorizers guide.
• Embedding Caching: Use Cache Embeddings to store generated vectors and reduce the cost of repeated embedding generation.

Drive user engagement using vector-based recommendation patterns:

• User Similarity: Find similar users or items using vector search.
• Real-Time Ranking: Combine vector similarity with metadata filtering and reranking to provide highly relevant results.
• Multi-Signal Matching: Use MultiVectorQuery to search across multiple embedding fields simultaneously.

Manage how data is stored and how to evolve your schema:

• Storage Formats: Choose between Hash and JSON formats, specifically considering how to handle nested data.
• Index Migration: Use the migrator helper, wizard, plan, apply, and validate workflow to move indices.
• Advanced Migration: Perform vector quantization with support for crash-safe resume, rollback, and a guided wizard flow.

RedisVL supports advanced querying capabilities:

• Complex Filtering: Combine multiple filter types including tag, numeric, geo, and text filters in a single query.
• Advanced Query Types: Implement hybrid search, multi-vector search, range queries, and specialized text queries.
• SQL Translation: Use the Write SQL Queries for Redis functionality to translate standard SQL syntax into Redis-compatible query syntax.

RedisVL provides several vectorizer classes to convert text into embeddings using various providers.

Supported Providers

• HuggingFace: HFTextVectorizer
• OpenAI: OpenAITextVectorizer
• Azure OpenAI: AzureOpenAITextVectorizer
• Google Cloud: GoogleGenAIVectorizer (Recommended) and VertexAIVectorizer (Deprecated)
• Cohere: CohereTextVectorizer
• AWS Bedrock: BedrockVectorizer
• Voyage AI: VoyageAIVectorizer
• Mistral AI: MistralAITextVectorizer
• Custom: CustomVectorizer for user-defined logic.

Improve the performance and relevance of your vector search:

• Reranking: Use cross-encoders and rerankers to improve the relevance of search results after the initial retrieval.
• Index Optimization: Use SVS-VAMANA for graph-based vector search with compression to optimize index performance.

RedisVL MCP (Model Context Protocol) is an MCP server that exposes existing Redis Search indexes as a stable, deterministic tool interface for AI applications and agent frameworks.

Use RedisVL MCP when:

• You want multiple AI assistants to share a single, approved retrieval surface.
• You want search behavior (like vector vs. fulltext) to be fixed by deployment configuration rather than client-side logic.
• You need a controlled read-only or read-write boundary for AI agents.
• You want to reuse existing Redis indexes without reimplementing query logic in every client.

Use direct RedisVL Python calls when:

• Your application needs to own the index lifecycle (creation/deletion).
• You need to construct complex, dynamic queries manually.
• You require full access to all RedisVL features directly in Python.