Google Gemini (Rest API) Integration

[TuanaCelik]

Summary and motivation

Google Gemini isn't only provided via Vertex AI. Let's create an integration for the REST API offering, too. There you only need a simple API Key - no need for a GCP account.

Checklist

If the request is accepted, ensure the following checklist is complete before closing this issue.

• [x] The code was merged in the main branch
• [x] Docs are published at https://docs.haystack.deepset.ai/
• [x] There is a Github workflow running the tests for the integration nightly and at every PR
• [x] A label named like integration:<your integration name> has been added to this repo
• [x] The labeler.yml file has been updated
• [x] The package has been released on PyPI
• [x] An integration tile has been added to https://github.com/deepset-ai/haystack-integrations
• [x] The integration has been listed in the Inventory section of this repo README
• [x] There is an example available to demonstrate the feature
• [x] The feature was announced through social media

[silvanocerza]

To support Gemini through Makersuite we will probably need to use a different library from the one used in Vertex AI.

We'll need google-generativeai, at least version 0.3.1. This because the google-cloud-aiplatform library used in the Google Vertex AI integration doesn't support Makersuite. Documentation is over here. Library repo here. Here to get the API key instead.

Also note that Makersuite is only available in the USA.

[TuanaCelik]

Would the ideal name for this integration be gemini-haystack or makersuite-haystack or something else? Wdyt?

[TuanaCelik]

Expanding on the comment above: especially if we want to document all of these integrations (which we do), we might end up wanting to rethink how we name them. E.g., in the google-vertexai-haystack integration we have the GeminiGeneroator, if this is to be a separate integration here, we probably want to start naming them to be explicit in how they're using Gemini??

E.g.: VertexAIGeminiGenerator and MakersuiteGeminiGenerator - I may be well off here.. Lmk what you think

[TuanaCelik]

Here's the draft PR when someone from core is back to finalize it: https://github.com/deepset-ai/haystack-core-integrations/pull/156

It has an example notebook in there. Works with:

• gemini-pro for just tex
• gemini-pro-vision for text + image
• GeminiChatGenerator for function calling

Needs:

• probably lots of name/package name changes to distinguish from the vertex counterpart
• tests :) Leaving it in draft in case you'd rather bin it

[TuanaCelik]

@dfokina @silvanocerza Let's get the docs going. Feel free to assign to me if needed. I can also take the example/demonstration. @bilgeyucel or @annthurium do one of you have room for the integration page? Or to switch the demo with me?

[dfokina]

@TuanaCelik @silvanocerza Docs are live: https://docs.haystack.deepset.ai/v2.0/docs/googleaigeminichatgenerator & https://docs.haystack.deepset.ai/v2.0/docs/googleaigeminigenerator