Hi , Please help in acheiving my AIM, as explained below with reason. AIM : want to have 1 Vector D

Hi , Please help in acheiving my AIM, as explained below with reason.

AIM : want to have 1 Vector DB per User .. for this design, plan to have created 50000 Vectorize Bindings (Upper Limit by CF) to the Worker (in first deployment) .. then, as new user signs-up pick a deterministic allocation of 1 Binding from "available pool of bindings" ..

What after 50000 users signed up ? .. create a new CF account and somehow find a way to direct the traffic based on a KV Map which keeps track for Worker Account and Binding ID with-in that worker, allocated for the user.

Question : is there a smarter way to acheive the AIM ?

Reason for My Ugly Design : Upfront Heavy Cost Incurred, for "just ONE" query done .. AFTER "FIRST" query, the incremental costs for query are very minimal .. so, if there are many NAMESPACE seperated users in ONE Vectorise Index .. Just ONE Active user will baloon-up the bill for rest dormant users (who never queries DB in that month)
This will result in ONLY ONE PAID User, while Vectorize will costing me for ALL Users in that Index.

Example :
CF Pricing : ((queried vectors + stored vectors) * dimensions * ($0.01 / 1,000,000)) + (stored vectors * dimensions * ($0.05 / 100,000,000))
Assume : stored vectors = 4200000, dimension = 1024, duration = 1 month
Case-1 : ZERO QUERIES , Total Cost = $2.1504
Case-2 : THREE QUERIES, Total Cost = $45.15

IiLearn Hi , Please help in acheiving my AIM, as explained below with reason. AIM : wan...

Elliot Hesp•7/24/25, 10:11 PM

You'd be better using workers for platforms and dynamically binding a db

EElliot Hesp You'd be better using workers for platforms and dynamically binding a db

iLearnOP•7/24/25, 10:13 PM

Does Workers for Platforms have upper limits ? Can you please point to web-page which can help ? Thanks.

Elliot Hesp•7/24/25, 10:15 PM

Probably the same limits. You'd most likely be able to reach out to support to request an increase in max databases. Although if you have 5000 bound to a worker and they're not actually being used, it's unlikely to be a valid request

EElliot Hesp Probably the same limits. You'd most likely be able to reach out to support to r...

iLearnOP•7/24/25, 10:20 PM

Yeah true and also expected as such request will be an abuse. But, how to design per-user-DB .. as my architecture solves per-user problem beautifully with DO .. but, got stuck to assign those users private and "cost-effective" Vector DB ..

EElliot Hesp You'd be better using workers for platforms and dynamically binding a db

iLearnOP•7/24/25, 10:27 PM

Just checked with ChatGPT for this. Following is response from ChatGPT. Please confirm, if your were referring in some other context. Thanks.

The statement: "using Workers for Platforms you can dynamically bind a Vectorize DB per user, which otherwise is not possible using Workers Paid plan" is not accurate.

Cloudflare Workers—whether on Free, Paid, or Platforms—use static bindings defined in your wrangler.toml (or dashboard) configuration. You must specify each Vectorize index you want to connect to at build/deploy time, not dynamically at runtime

IiLearn Just checked with ChatGPT for this. Following is response from ChatGPT. Please c...

Isaac McFadyen•7/24/25, 10:53 PM

ChatGPT is incorrect. Workers for Platforms allows you to dynamically create Workers (and therefore bind Vectorize DBs to those Workers) via an API.

IIsaac McFadyen ChatGPT is incorrect. Workers for Platforms allows you to dynamically create Wor...

iLearnOP•7/25/25, 8:08 PM

Thanks for correcting the understanding.

Unfortunately, this also doesnot solve my AIM (= "Isolated private Vectorize DB "per User"), as i shall be hitting the limit
"Vectorize Indexes per account 50,000 (Workers Paid)"

Q#1 : How to scale after 50k (PROGRAMATICALLY) ?
Q#2 : IF, Q#1 is NOT possible, is Cloudflare Infra "capable" to handle Vectorize DB "per User", if manual request need to be dropped to support to increase the limits, as i start nearing the 50k limit ?

Thanks in advance. These Qs are very important for me to architect the whole AI pipeline with serverless approach for "planet earth". (DO solves one part by "storing chunks" for planet-earth, but Vectorize is bottle neck for full fledged AI pipeline).

EElliot Hesp You'd be better using workers for platforms and dynamically binding a db

iLearnOP•7/28/25, 4:51 PM

Thanks. Issac also confirmed the same.

KalmonJ•7/29/25, 1:10 PM

VECTOR_QUERY_ERROR (code = 10000): Authentication error

Vivek•7/30/25, 5:56 PM

hello everyone I was preparing data on estimated cost for vectorize for my company can you guys tell if I am correct here with the calculations

Bethink•8/4/25, 4:37 PM

Let’s say you use Vectorize to index news media. What is the best way to periodically expire vectors?

Dev•8/8/25, 4:06 AM

when will --experimental-vectorize-bind-to-prod--experimental-vectorize-bind-to-prod flag get stable, it fails 4/5 times resulting in 10000 Auth error

iLearnOP•8/17/25, 5:36 PM

Please confirm following statement :
"A vector can only be part of a single namespace .."
Source : https://developers.cloudflare.com/vectorize/reference/metadata-filtering/#namespace-versus-metadata-filtering

My understanding and Correct Statement should be : "A vector IDID can only be part of a single namespace .."

Is following code "sampleVectors_1" correct ? : (If not, plz share corrected version/example)

embedding_values = [32.4, 74.1, 3.2, ...];

//sampleVectors_1 : (Same embedding value, same namespace, only Vector-ID differ)

const sampleVectors_1: Array<VectorizeVector> = [
  {  id: "1",  values: embedding_values,  namespace: "text", },
  {  id: "2",  values: embedding_values,  namespace: "text", },
]

embedding_values = [32.4, 74.1, 3.2, ...];

//sampleVectors_1 : (Same embedding value, same namespace, only Vector-ID differ)

const sampleVectors_1: Array<VectorizeVector> = [
  {  id: "1",  values: embedding_values,  namespace: "text", },
  {  id: "2",  values: embedding_values,  namespace: "text", },
]

Cloudflare Docs

Metadata filtering

In addition to providing an input vector to your query, you can also filter by vector metadata associated with every vector. Query results will only include vectors that match the filter criteria, meaning that filter is applied first, and the topK results are taken from the filtered set.

yinxingmaiming6409•8/25/25, 8:45 AM

When can a new version be released again, with query speed and over 5 million data?

garvitg•8/27/25, 3:11 PM

Vectorize update:

2025-08-27: Vectorize now provides the ability to list all the vector identifiers in your index using the list-vectors operation: https://developers.cloudflare.com/changelog/2025-08-26-vectorize-list-vectors/.

Cloudflare Docs

List all vectors in a Vectorize index with the new list-vectors ope...

Enumerate all vector identifiers in your Vectorize indexes for bulk operations, auditing, and data migration.

zsueo•8/28/25, 12:26 PM

are the new vectorize defintions usable? wrangler types seems to only use VectorizeIndex but im noticing that doesn't cover the full api mentioned by the docs, and there is an abstract class Vectorize which has full coverage

zsueo•8/28/25, 8:25 PM

nvm figured it out

Sonjj•8/30/25, 4:35 AM

The Autorag feature for websites does not show all domains currently used on cloudflare. Can anyone help me with this? Thanks

Ggarvitg 📢 Vectorize update: 2025-08-27: Vectorize now provides the ability to list al...

yinxingmaiming6409•9/1/25, 6:42 AM

Can only be operated with the command 'wrangler'? No worker API?

Ggarvitg 📢 Vectorize update: 2025-08-27: Vectorize now provides the ability to list al...

nclevenger•9/2/25, 12:36 PM

Does list vectors have a workers API? The docs only show the REST API

vijay•9/3/25, 2:03 PM

Hey everyone, I’ve been exploring Vectorize for my product in a multi-tenant setup where each tenant gets their own index chunk via namespaces. Because of the worker binding count limit (and lack of dynamic creation), I went with the REST API approach.

But I recently discovered that Cloudflare rate limits HTTP requests (1,000 per 5 minutes), which makes me wonder how Vectorize is supposed to be used effectively in a multi-tenant setup.

I’m confused because Cloudflare advertises 50k indexes per account, but in practice this doesn’t seem possible due to the worker binding limit + request rate limits. Feeling a bit stuck and frustrated—what’s the recommended approach here?

zhichenguo•9/4/25, 5:13 PM

Hey all, I am using

vitest-pool-workers

vitest-pool-workers

to test my app in Github Action, and running into the following error with

VectorizeIndex.query()

VectorizeIndex.query()

workerd/io/worker.c++:2169: info: uncaught exception; source = Uncaught (in promise); stack = Error: VECTOR_QUERY_ERROR: Status + 500
    at VectorizeIndexImpl._send (cloudflare-internal:vectorize-api:185:23)
    at VectorizeIndexImpl.queryImplV2 (cloudflare-internal:vectorize-api:205:21)
    at VectorizeIndexImpl.query (cloudflare-internal:vectorize-api:45:20)

workerd/io/worker.c++:2169: info: uncaught exception; source = Uncaught (in promise); stack = Error: VECTOR_QUERY_ERROR: Status + 500
    at VectorizeIndexImpl._send (cloudflare-internal:vectorize-api:185:23)
    at VectorizeIndexImpl.queryImplV2 (cloudflare-internal:vectorize-api:205:21)
    at VectorizeIndexImpl.query (cloudflare-internal:vectorize-api:45:20)

It works when testing locally locally but fails in the Github Action environment. Has anyone else ran into this issue?
Thanks in advance! Please tell me if I should provide any more details

Nnclevenger Does anyone know what this error means: `VECTOR_QUERY_ERROR: Status + 500`? I w...

zhichenguo•9/4/25, 9:44 PM

I haven't found any references to the

VECTOR_QUERY_ERROR: STATUS + 500

VECTOR_QUERY_ERROR: STATUS + 500

error other than this message and this thread

I am getting 500 when I try and insert a - Cloudflare Developers

I am getting 500 when I try and insert a new vector record: Error: VECTOR_INSERT_ERROR: Status + 500

matt_breakfall•9/5/25, 9:40 PM

Is there any significance to the mutationIds? i.e. c0e522b1-3541-4aa8-a298-32c5c5fbf5b9. All i want to know is if there are in any kind of order. It's great seeing the latest one in status but which ones came inbetween? Similiar vector databases return a mutation number in a sequence]

Rapid•9/6/25, 6:13 PM

Hiya, did anyone create face recognition using vectorize? would love to see examples (I got something working, but pretty obvious matches get a 0.35~ score

Vlad Lykhonis•9/24/25, 9:39 AM

Are these limits hard or can be raised later? https://developers.cloudflare.com/vectorize/platform/limits/

Cloudflare Docs

Limits

The following limits apply to accounts, indexes and vectors (as specified):

Anon•10/4/25, 3:07 AM

What does this cost???

Zetax•10/4/25, 4:29 AM

Not sure what I'm doing wrong, but each time I try to query for embeddings with metadata filters, it gives me no results, even if the embeddings are the same. First image is the result without filters (see the userId metadata key?)

This is my code:

export const search = base
    .route({ method: 'POST', path: '/ai/vectors/search' })
    .input(
        z.object({
            embeddings: z.array(z.number()),
            topK: z.number().optional().default(10),
            filter: z.record(z.string(), z.any()).optional(),
        })
    )
    .output(vectorizeMatchesSchema)
    .handler(async ({ input, context }) => {
        const options: VectorizeQueryOptions = {
            returnMetadata: true,
            topK: input.topK,
        };

        if (input.filter) {
            options.filter = input.filter;
        }

        const results = await context.env.VECTORIZE.query(input.embeddings, options);
        return vectorizeMatchesSchema.parse(results);
    });

export const search = base
    .route({ method: 'POST', path: '/ai/vectors/search' })
    .input(
        z.object({
            embeddings: z.array(z.number()),
            topK: z.number().optional().default(10),
            filter: z.record(z.string(), z.any()).optional(),
        })
    )
    .output(vectorizeMatchesSchema)
    .handler(async ({ input, context }) => {
        const options: VectorizeQueryOptions = {
            returnMetadata: true,
            topK: input.topK,
        };

        if (input.filter) {
            options.filter = input.filter;
        }

        const results = await context.env.VECTORIZE.query(input.embeddings, options);
        return vectorizeMatchesSchema.parse(results);
    });

    return await client.ai.vectors.search({
      embeddings: embeddings.data[0] ?? [],
      topK,
      filter: {
        userId: {
          $eq: userId,
        },
      },
    });
  }

    return await client.ai.vectors.search({
      embeddings: embeddings.data[0] ?? [],
      topK,
      filter: {
        userId: {
          $eq: userId,
        },
      },
    });
  }

Bildschirmfoto_2025-10-04_um_06.26.31.png

Zzsueo nvm figured it out

Austin Williams•10/20/25, 1:48 PM

Running into the same thing. Mind sharing how you resolved the type issues?

AAustin Williams Running into the same thing. Mind sharing how you resolved the type issues?

zsueo•10/20/25, 1:50 PM

Uhh from what I can remember I vectorizeIndex does cover the full apo and I was wrong, I'm not sure what the Vectorize class was in worker types tho

Zzsueo Uhh from what I can remember I vectorizeIndex does cover the full apo and I was ...

Austin Williams•10/20/25, 1:53 PM

Weird,

VectorizeIndex

VectorizeIndex

looks like the old types for me. For example,

upsert

upsert

is typed to return

Promise<VectorizeVectorMutation>

Promise<VectorizeVectorMutation>

which doesn't include the

mutationId

mutationId

zsueo•10/20/25, 1:55 PM

Could you drop the d.ts file? I might remember more if I have more context

Zzsueo Could you drop the d.ts file? I might remember more if I have more context

Austin Williams•10/20/25, 1:58 PM

Here's the Vectorize related stuff. @garvitg would appreciate any insight here

message.txt10.5KB

carter•10/27/25, 12:49 AM

Hey all - are there plans to extend the Vectorize query API to either
1) increase per-query limits (from 20/100+), or
2) add pagination/offsets

Thinking of an interface like in https://qdrant.tech/documentation/concepts/search/#pagination
Thanks!

Search - Qdrant

Qdrant is an Open-Source Vector Database and Vector Search Engine written in Rust. It provides fast and scalable vector similarity search service with convenient API.

Ccarter Hey all - are there plans to extend the Vectorize query API to either 1) increas...

carter•10/27/25, 12:50 AM

Curious about the 2nd point in particular bc it's more of a wrapper/QoL feature

piyush534•10/28/25, 5:41 PM

Hi I have just migrated my current setup from qdrant to vectorize but I m facing several issues basically in large pdf it is not retrieving relvant queries can anyone who has a knowledge in vectorize rag flow can help me ..

Shoro AI•10/29/25, 12:44 AM

@zsueo @garvitg - Can u pls help on above query of Piyush

AAiden Looks like specifically editing vector DBs where we get the 9002 internal error

Aiden•11/5/25, 11:05 PM

Marak•11/7/25, 6:33 PM

Hello. Is it possible to delete vectors by namespace? We have found

deleteByIds

deleteByIds

, is there a way to delete all vectors that match a namespace?

Marak•11/7/25, 6:40 PM

Am I to understand the work-around for deleting vectors by namespace is to run a dummy query ( to match all results ) and loop batch delete by ID using

deleteByIds

deleteByIds

MMarak Am I to understand the work-around for deleting vectors by namespace is to run a...

yevgen•11/10/25, 9:30 AM

Correct. That's the workaround that exists today. We plan to support query filtering for delete operations in H1 2026.

Yyevgen Correct. That's the workaround that exists today. We plan to support query filte...

Marak•11/10/25, 9:32 AM

Hello. Appreciate that buddy. Thank you. We were able to implement the work-around and are pleased with the results. Vectorize is very nice.

cosbgn•11/11/25, 8:56 AM

Is there a way to delete a namespace?

cosbgn•11/11/25, 8:57 AM

e.g. I have one namespace per customer. When a customer delete its account, should I query all vectors and use deleteById or there is a way to simply delete a namespace?

cosbgn•11/11/25, 8:58 AM

and even if I use deleteById, will I then have a hunging empty namespace which uses one of my 5000 max namespace quotas or vectorize will automatically free it?

cosbgn•11/11/25, 8:59 AM

So I just read the message above and

deleteById

deleteById

seems the way, but will the empty namespace still count against the 5K quota?

ElishaKay•11/12/25, 9:53 AM

Do I always need to deploy a Cloudflare Worker in order to access my Cloudflare Vectorize stores?

Or can I access my Cloudflare Vector Database from an Langchain or Langgraph Script?

Same question for the AutoRAG - it looks like I need a worker for the AI Search feature

I'm looking at this python library of Cloudflare Vectorize

Docs by LangChain

Cloudflare Vectorize - Docs by LangChain

Cloudflare API | Vectorize

Interact with Cloudflare's products and services via the Cloudflare API

Vivek•11/17/25, 8:06 PM

Has anyone experienced a vector count mismatch with LangChain + Cloudflare Vectorize?
I'm noticing a discrepancy where my D1 document count doesn't match my Vectorize index. For example, I have 12 raw documents in D1, but 48 vectors in Vectorize (4x the amount). Also I am seeing this discrepency in other indices where the vectors are 4x the raw chunks in D1.

Have others run into this issue? Any insights on what might be causing this would be appreciated!

Victor•11/18/25, 9:48 PM

Is the 5mil limit of vectors per index a soft limit that can be raised or hard limit?

Dmitri•11/19/25, 4:26 PM

Could anybody advice - how to get help or support with vectorize? I don't see any direct support option anywhere and community shows only some very old topics in random categories. It feels like index stopped updating yesterday during outage. It allows upsert, but won't return any results. And processedUpToDatetime shows yesterday - while system upserted thousands of vectors in last 24 hours.

Hi , Please help in acheiving my AIM, as explained below with reason. AIM : want to have 1 Vector D

Similar Threads

Similar Threads

Similar Threads