RAG not working with Anthropic Key and JS-SDK

Support
,
1

Hi!

I tried to leverage the full RAG-functionality of weaviate-cloud. I could not get any results with the JS-SDK, using my anthropic API-key in the respective header. Instead I received the following Error:

WeaviateQueryError: Query call with protocol gRPC failed with message: /weaviate.v1.Weaviate/Search UNKNOWN: explorer: get class: concurrentTargetVectorSearch): explorer: get class: extend: extend generate: client not found, empty provider
    at <anonymous> (/xxx/node_modules/.pnpm/weaviate-client@3.4.0_encoding@0.1.13/node_modules/weaviate-client/dist/node/esm/grpc/searcher.js:46:19)
    at process.processTicksAndRejections (node:internal/process/task_queues:105:5)

Apart from that I could easily perform embeddings and hybrid search, so RAG seems the culprit here. I tried both, the REST and the grpc Endpoint, because the documentation does not clarify which one to use. Using the grpc endpoint resulted in this error:

WeaviateStartUpError: Weaviate startup failed with message: Weaviate failed to startup with message: The request to Weaviate failed with status code: 415 and message:
    at <anonymous> (/xxx/node_modules/.pnpm/weaviate-client@3.4.0_encoding@0.1.13/node_modules/weaviate-client/dist/node/esm/connection/helpers.js:52:11)

Can somebody tell me what’s going on here?

Thank you!

2

Hey there, could you share snippets of your connection function (with headers included) and the code that produces these errors? What Weaviate and JS client version are you using?

3

Hi,

client version: 3.4.0,
weaviate cloud database version(sandbox): 1.29.0,

I connect with this call:

 this.client = await weaviate.connectToWeaviateCloud(this.url, {
 authCredentials: 'XYZ',
      headers: {
        'X-Anthropic-Api-Key':
          'XXX',
      },
    });

the throwing code looks like this:

const results = await this.client.collections
      .get('collection')
      .withTenant('tenantX');.generate.nearText(
      'query',
      {
        groupedTask: 'xxx',
      },
      {
        limit: 10,
      },
    );

Thank you for looking into this!

4

From the error you have, it looks like you might have misconfigured your collection.

Did you include the generative anthropic module or a vectorizer module??

The snippet you shared (minus the “;” before generate) worked for me with this collection configuration.

await client.collections.create({
      name: 'Collection',
      // Define your vectorizer and Anthropic generative model
      vectorizers: weaviate.configure.vectorizer.text2VecOpenAI({
        sourceProperties: ['title','text']
      }),
      generative: weaviate.configure.generative.anthropic(),
      multiTenancy: weaviate.configure.multiTenancy({enabled: true, autoTenantCreation: true})
    });

Let me know if any of these fix the issue for you.

5

Sorry for the ;! That’s just a typo originating in me trying to generalise my code for this thread.

The vectorizers-property was configured as suggested by you, but I actually missed that I have to configure my collection to use a generative model too. My bad, thank you very much for bringing that up.

Maybe a more prominent hint on collection-configuration in the Rag Documentation and a more verbose ErrorMessage would improve the DX here!

1 Like
6

I figured, i only pointed it (;) out incase that was an error waiting to happen :wink:

Happy it worked, thanks for the feedback on the docs, we do have a drop down pointing users to a page that details the required config for RAG, do you think we could do more to make it more obvious?

7

Maybe just pin a minimal collection config on top with all properties that are required for RAG.

But an error message stating that the collection misses a generative AI model would be most helpful.

Thank you :slight_smile:

1 Like