Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- Mistral listed its hosted Search Index as a Studio public preview on October 30, 2026. Mistral hosts the index storage for document retrieval in agents and workflows.
- Creating an index requires an API key from the workspace that holds it and the create_search_indexes permission. Choose an index name, embedding model, and document schema.
- Documents are split into chunks, embedded, and sent with their fields. Retrieval supports vector, keyword, or hybrid search with typed-field filters.
- Index creation and schema changes are asynchronous; wait until the index is ready before using it. Search Toolkit can connect to other backends and is distinct from the hosted Search Index.
What did the October 30 preview announce?
Mistral's October 30, 2026 release notes list the managed search index as a Studio Public Preview and Mistral-Hosted feature. Developers can configure a hosted vector index with a name, an embedding model, which turns text into numeric representations for semantic matching, and an optional document-field schema, then ingest and search through one API in the Mistral SDK without deploying a search backend themselves.
The overview describes Search Index as storage for searchable document chunks, so agents and workflows can retrieve company information as answer context. It remains in public preview: its API, behavior, and tooling may change, and an update may require code changes.
How Search Index differs from Search Toolkit
Search Index is Mistral's hosted search backend. Search Toolkit provides a Python client and components for extraction, splitting, embedding, and retrieval. Mistral says the toolkit can connect to other backends too, so using the toolkit does not require storing an index in Mistral's hosted service.
Decide whether you want Mistral to host the storage, then decide whether to use the toolkit to organize ingestion and retrieval. If you already have a backend or pipeline, the official pages do not require you to migrate first; check that your backend and current tools fit the workflow you need.
Check workspace access and fields before creating
Creating an index requires an API key from the workspace that holds the index and the create_search_indexes permission in that workspace. Confirm the key's workspace and permissions first. If an index is not visible, check the key's workspace and your permission on that index instead of assuming it belongs to another account.
Creation asks for an index name, embedding model, and document schema. Custom document and chunk fields can support structured filters, such as product, category, or publication date. Decide which fields you need to retrieve or filter on before designing the schema, rather than leaving the document structure until after ingestion.
Prepare documents and choose a retrieval method
The documented flow adds documents, splits their content, creates embeddings, and sends the chunks with their fields. Start with a small, representative set of authorized material and check which fields survive chunking before increasing the import. The official pages do not publish capacity or rate numbers that apply to every account.
Retrieval supports vector, keyword, and hybrid search, with field filters. Retrieved passages can provide context to an agent or workflow. Match the method to the query and metadata you have; the documentation does not claim that one search mode is more accurate for every collection.
Wait for readiness and verify preview eligibility
Index creation and schema changes are asynchronous. A successful request does not mean an index is ready to query. Check its status with the current interface and wait for ready before searching. Schema updates also take time, so do not treat an in-progress index as already using its new fields.
These two official pages mark Search Index as a Public Preview, but do not specify pricing, regions, complete account eligibility, or a fixed service commitment. Verify those conditions in your current workspace and the official console before planning a rollout or budget; the preview label alone does not establish that every account, region, or usage level is the same.
Sources
Frequently Asked Questions
Do the 3 or 10 Managed Deployments slots also limit Search Index?
Do not transfer that quota between services. Managed Deployments runs workflow workers; Search Index provides document retrieval storage. The Search Index pages reviewed here do not state that quota. See the <a href="/en/resources/mistral-managed-deployments-github-quota-guide">Managed Deployments guide</a> for workflow hosting eligibility and limits, and confirm index quotas separately.
Does a hosted index accept every file format directly?
That is not established. These pages describe extraction, splitting, embedding, and retrieval components without a complete file-format support list. Confirm support in the current Toolkit or your chosen processing pipeline before importing material.
Does retrieving a relevant passage prove the agent’s final answer is correct?
Retrieved passages provide context to an agent or workflow. Check whether the answer follows that evidence, preserves its conditions, and uses information that remains current. The reviewed documentation provides no answer-accuracy guarantee covering every collection.