Niches · Retrieval & vector search
Store a fact and find it again at once, then change or delete it
Two or three companies' docs show a full fit.
What counts as done
A record written in one call is returned by a matching query in the next call, an update or delete by id is reflected in the next query, and old versions do not crowd results.
Measured, for our own tests: a written record is searchable within 1 second. No docs we read state these figures, so they don't count against a listing here.
How agents ask for it: remember this for later
, what did I say about X
, update that note
, forget that preference
Does the job 2
The docs show it meets the test above, and nothing we read says otherwise.
| Listing | Why |
|---|---|
| LanceDB B LanceDB Systems, Inc. | With the real-time engine (Embedded and Enterprise) writes are vector-searchable within milliseconds, and delete and update take a predicate. source |
| Milvus and Zilliz Cloud API + MCP B Zilliz Inc. | Strong consistency makes a write visible to the next search, and upsert and delete by id or filter are in the API; the MCP server has insert and delete_entities. source |
Does the job, with a limit 11
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. | Update and delete by id work in the API and in chroma-mcp, but #4631 reports metadata-only updates running extremely slowly, and no write-visibility delay is stated. source |
| Cloudflare AI Search BB Cloudflare, Inc. | The Items API uploads and deletes documents, but indexing runs as a job, so a written record is not shown to be searchable in the next call. source |
| Elasticsearch C Elasticsearch B.V. | Index, update and delete requests take refresh=wait_for, which replies only once the change is visible to search, but the agent must set it on each write, and update or delete by query was not on the pages read. source |
| Epsilla Vector Database E Epsilla, Inc. | Upsert and delete by primary key exist, but the docs do not state when a write becomes searchable. source |
| Meilisearch A Meili SAS | Document adds, updates and deletes are enqueued as tasks, so a record is not searchable in the very next call unless the agent first polls the task to succeeded. source |
| OpenSearch BB OpenSearch Software Foundation, a project of The Linux Foundation | The MCP server's agentic memory tools add, update and delete memories by id or query, but they need an admin-created memory container, and write-visibility timing was not read. source |
| Pinecone API + MCP A Pinecone Systems Inc. | Pinecone is eventually consistent with a slight delay before writes show; the agent can compare LSNs to confirm, but there is no wait or strong-read option on the write. source |
| Qdrant API + MCP BB Qdrant Solutions GmbH | The REST API takes wait=true and updates or deletes by id or filter, but the MCP server has only store and find, so changed facts pile up as old versions (#74). source |
| Typesense API + MCP BB Typesense, Inc. | Documents can be upserted, partially updated and deleted by id, but the pages read do not state when a write becomes searchable. source |
| Upstash Vector API + MCP B Upstash, Inc. | Update edits one record in place with a metadata patch mode, and deletes take id, prefix or filter, in the API and the hosted MCP server, but write visibility timing is not stated on the pages read. source |
| Weaviate API + MCP C Weaviate B.V. | The API updates and deletes objects, but the MCP server's only write tool is weaviate-objects-upsert, off by default, with no delete; write-visibility timing on a single node is not stated on the pages read. source |
Couldn't check 1
The docs were closed to us, or silent. These carry no mark either way.
| Listing | Why |
|---|---|
| Vespa BB Vespa.ai AS | Document API update and remove are documented, but when a write becomes searchable was not on the pages read. |
What stops agents finishing
- eventual consistency with no wait or strong-read option
- MCP server has no update or delete tool
Around it
Before: agent-memory.store-and-confirm. After: agent-memory.user-facts-recall.
Where this comes from
Weight 10 of 100 in Retrieval & vector search: 4 of 37 accounts (Qdrant MCP store and find, no delete tool, metadata updates), plus the update-delay buyer check in the category context
The same as data: /api/v1/use-cases/vector-search.memory-write-read.json, or the MCP tool get_use_case with id vector-search.memory-write-read. 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.