Introducing Search Toolkit

Mistral AI announced the release of Search Toolkit in public preview. According to the company, Search Toolkit is a composable framework for building production search pipelines for AI applications. It is open source and runs on cloud, on-premises, and edge infrastructure.

Mistral states that teams building search infrastructure spend significant engineering time stitching together separate tools for ingestion, retrieval, and evaluation. Search Toolkit brings these three into a single framework with a shared interface, which Mistral says allows teams to focus on search quality rather than maintaining integrations. The framework addresses enterprise search, RAG and retrieval quality, and domain-specific retrieval, according to the announcement.

Search Toolkit includes ingestion, retrieval, and evaluation modules. The ingestion module processes data from multiple sources with configurable pipelines, handling document parsing, chunking, and embedding generation, with custom formats supported through a standard adapter interface. Retrieval ships with BM25 sparse retrieval, dense embedding-based retrieval, and hybrid configurations. Evaluation provides built-in metrics: recall, precision, MRR, and NDCG, run against users' own test sets.

The announcement says the framework is designed for agentic use, allowing agents to perform semantic search on an index and pull live data from source systems through Connectors using MCP integrations.

Mistral says Search Toolkit has been battle tested across financial services, manufacturing, public sector, and media and entertainment. CMA CGM uses Search Toolkit alongside Voxtral to help journalists detect fake news; the pipeline processes audio from three distinct data sources and returns alerts within 15 seconds end to end, according to Mistral.

A starter app template is available, requiring Docker and uv. It includes pre-configured Vespa indexing, hybrid retrieval (BM25 plus vector), and a sample data ingestion pipeline.