Niches · Retrieval & vector search
Search inside one tenant, user or metadata slice
Four or more companies' docs show a full fit.
What counts as done
Every result matches the filter, the result count equals the requested k when k matches exist, and no other tenant's records appear.
Measured, for our own tests: p95 latency with a filter matching 1 per cent of records is within twice the unfiltered p95. No docs we read state these figures, so they don't count against a listing here.
How agents ask for it: search only this customer's documents
, results from 2025 only
, only from the HR folder
, this user's memories
Does the job 4
The docs show it meets the test above, and nothing we read says otherwise.
| Listing | Why |
|---|---|
| Meilisearch A Meili SAS | Results are first filtered, then ranked within the filtered set, and tenant tokens restrict each user's searches to their own records. source |
| Milvus and Zilliz Cloud API + MCP B Zilliz Inc. | Filtered search applies metadata filtering before the ANN search, so limit comes from matching entities, and multi-tenancy is offered at database, collection, partition or partition-key level. source |
| Pinecone API + MCP A Pinecone Systems Inc. | The optional filter narrows candidates before ranking, so top_k comes from matching documents, and namespaces separate tenants in one index. source |
| Vespa BB Vespa.ai AS | nearestNeighbor combines with query filters and Vespa chooses pre- or post-filtering, and streaming mode searches one user's or tenant's subset without a shared index. source |
Does the job, with a limit 10
It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.
| Listing | Why |
|---|---|
| Chroma API + MCP D Chroma Inc. | Metadata where filters and tenant and database scoping exist, but #4089 reports where filters breaking queries over 20 million chunks on a Docker server, and pre- versus post-filtering is not stated on the pages read. source |
| Cloudflare AI Search BB Cloudflare, Inc. | Metadata filters apply before retrieval, and namespaces and folders separate content, but filters match only the first 64 bytes of a string, and no latency figure for selective filters is given. source |
| Elasticsearch C Elasticsearch B.V. | A filter inside the knn clause is applied during approximate search so k matching documents return, but #162043 found a nested-field pre-filter returning nothing, the use case's named blocker; it is now closed. source |
| Epsilla Vector Database E Epsilla, Inc. | query takes a boolean filter expression, but the docs do not say whether it applies before or after ranking, and no tenant or namespace feature was found. source |
| LanceDB B LanceDB Systems, Inc. | where(..., prefilter=True) filters before the vector search, but #3095 found prefilter inverted on hybrid queries (now closed as fixed), and tenant separation means separate tables. source |
| OpenSearch BB OpenSearch Software Foundation, a project of The Linux Foundation | Efficient k-NN filtering applies the filter during the vector search so k results return (Lucene HNSW, Faiss HNSW and IVF), but #203 (open) shows the MCP server ignoring filters in multi mode. source |
| Qdrant API + MCP BB Qdrant Solutions GmbH | Filters apply inside the HNSW search via payload indexes, with ACORN for strict filters and is_tenant partitioning, but #7211 reports recall dropping with multitenancy and #7210 high filtered latency at 4.6M points. source |
| Typesense API + MCP BB Typesense, Inc. | filter_by applies during the vector search and scoped search keys can embed a tenant filter, but #3159 shows a broad filter with a small intersection taking seconds, well past the twice-unfiltered latency bar. source |
| Upstash Vector API + MCP B Upstash, Inc. | Namespaces separate tenants, but filtering is in-filtering within a budget that falls back to post-filtering, and the docs say highly selective filters may return fewer than topK results. source |
| Weaviate API + MCP C Weaviate B.V. | Multi-tenancy stores each tenant on its own shard, invisible to others, and filters run with ACORN, but #7681 reports hybrid filters not acting as strict pre-filters. source |
What stops agents finishing
- filter applied after ranking so fewer than k results come back
- filter ignored in one tool mode
- nested-field filter returns nothing without an error
Around it
Before: agent-auth.resolve-user. After: vector-search.rag-over-docs.
Where this comes from
Weight 16 of 100 in Retrieval & vector search: 9 of 37 accounts, almost all failures (filters ignored, post-filtered, inverted, slow or returning nothing); weighted for how often filters decide correctness
The same as data: /api/v1/use-cases/vector-search.filtered-tenant.json, or the MCP tool get_use_case with id vector-search.filtered-tenant. Other jobs in this category: /api/v1/niches?category=vector-search.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.