Mistral: Search Toolkit for production search pipelines

Mistral has unveiled Search Toolkit, a composable framework for building production search pipelines for AI applications. The tool is open-source and, according to Mistral, “runs wherever your infrastructure runs” — in the cloud, on-premises or at the network edge.

What does Search Toolkit solve?

Search Toolkit unifies ingestion, retrieval and evaluation under a single shared interface. Instead of stitching together separate tools, the framework reduces the engineering burden by consolidating the entire search flow. A pipeline is a term for a series of connected processing steps that data passes through.

What are the main components?

The toolkit has three core components. Ingestion parses documents, splits content into chunks (chunking) and generates embeddings from multiple sources. Retrieval offers BM25 sparse retrieval, dense retrieval based on embeddings, and hybrid approaches. Evaluation measures quality with metrics such as recall, precision, MRR and NDCG on your own test sets.

What is it used for?

The main use cases are search within enterprises (wikis, support systems, repositories, code bases) and the optimization of RAG, where the quality of retrieval is separated from the quality of generation. The toolkit also targets domain-specific retrieval for legal, medical and financial documents, and agentic workflows. Integration is available through standard APIs and Vespa infrastructure for indexing.