Search Index

A Search Index stores searchable document chunks for your agents and workflows. Mistral hosts the storage; you choose the embedding model and the fields needed for search and filtering.

Use Search Index to ground answers in company documents without operating a vector database. Add documents, retrieve passages with vector or keyword search, and filter on typed fields such as product, category, or publication date.

The Search Toolkit provides the Python client and components for extraction, splitting, embedding, and retrieval. Search Index is the hosted backend for those pipelines; the toolkit can also connect to other backends.

1. Create an index. Choose a name, an embedding model, and a document schema.
2. Add documents. Split content into chunks, embed them, and send the chunks with their fields.
3. Retrieve results. Use vector search, keyword search, or hybrid retrieval with filters.
4. Use the results. Provide relevant passages to an agent or workflow as context.

Custom fields on documents and chunks let you combine semantic retrieval with structured filtering. Index creation and schema changes are asynchronous: wait for the index to be ready before using it.

• Quickstart: create and query an index: run an end-to-end Python example.
• Basic operations: define fields, create an index, ingest and search documents, update the schema, and delete data.
• Search Toolkit: build ingestion and retrieval pipelines.

Use an API key for your workspace. Creating an index requires the create_search_indexes permission. Check the workspace and your permissions if an index is not visible. Keep credentials out of source control.