Document how to reduce the amount of dependencies needed

[fire015]

Describe the bug
Installing llamaindex adds almost 900MB of packages to node modules! Ironically the workflow package is not installed and has to be done separately. Is there a way to reduce the total sizes of dependencies especially if just using OpenAI? I don't need Gemini etc but it seems to install every LLM provider by default.

To Reproduce

npm i llamaindex

• OS: macOS
• JS Runtime / Framework / Bundler (select all applicable)
• Node.js

[marcusschiesser]

You can reduce this by just installing the core package https://www.npmjs.com/package/@llamaindex/core and the LLM provider that you need (e.g. Openai: https://www.npmjs.com/package/@llamaindex/openai)

How about adding a "Reduce Dependencies" section to the docs: https://ts.llamaindex.ai/docs/llamaindex ?

"Guide" looks like a good place:
https://github.com/run-llama/LlamaIndexTS/blob/main/apps/next/src/content/docs/llamaindex/guide/meta.json

Anyone wants to help with that?

[fire015]

Unfortunately installing core alone is not enough to get this working. For example you can't access to vector stores they seem to be part of the main package.

[marcusschiesser]

Yes, if you need vector stores, then please wait for #1587 - coming soon!

[AndreMaz]

@marcusschiesser any ETA for this? I'm getting the following error at Vercel:

after bumping llamaindex from 0.6.9 to 0.8.31.

I think that importing the @llamaindex/core , @llamaindex/openai and @llamaindex/vector-store could reduce the bundle size of my (serverless) function

Link to Vercel docs: https://vercel.com/docs/functions/runtimes#bundle-size-limits
Related issue: #1572

[AndreMaz]

Turns out that, in my case, the culprit is the onnxruntime-node that takes 727.79MB of space

[marcusschiesser]

@AndreMaz which vector store are you using? We're already extracted postgres and azure. Using those with core should result in a small package size.

onnx is used by huggingface and chromadb. We will optimize bundle sizes and import handling in the 0.9 release

[AndreMaz]

Hey @marcusschiesser thank you for replying and thank you guys for your work!

I'm using PGVectorStore. Can you also take a look at #1598 ? At the moment I'm still relying on llamaindex to import VectorStoreIndex and serviceContextFromDefaults

[marcusschiesser]

@AndreMaz just released the 0.9 package, for migration check: https://ts.llamaindex.ai/docs/llamaindex/migration/0.8-to-0.9