New! Full Support for Google Vertex AI Gemini Pro and Pro Vision

[tzolov]

Find more about Spring AI - Vertex AI Gemini:

• VertexAI Gemini Chat
• Gemini Function Calling
• Support of Multimodal chat requests
• Native AOT support

[DenisStrazhevich]

Great job! @tzolov Can you please tell me if there are plans to integrate Google AI to work with models (e.g. Gemini) without a Google Cloud account, but with an API key?

[tzolov]

This is a great question @DenisStrazhevich and perhaps you can help me with it a bit.
I've added already support passing the credential JSON file as a spring.ai.vertex.ai.gemini.credentialsUri, that when provided are used to create

GoogleCredentials credentials = GoogleCredentials.fromStream(connectionProperties.getCredentialsUri().getInputStream());
return new VertexAI(connectionProperties.getProjectId(), connectionProperties.getLocation(), credentials);

In the VertexAiGeminiAutoConfiguration#vertexAi.

However I wan't able to property verify this as I failed to extract the right credentials for my gemini account.

May you know (or have understandable docs) how to generate the credentials files and eventually the api-key from an existing service?
I'm a bit lost in google's console and studio :/ But if i figure out how to obtain those for my test account it will be trivial to implement the authentication.

[DenisStrazhevich]

@tzolov I think it should help. It describes how to generate a credentials.json file in your Google Cloud console.

But my question was a little bit about something else, I put it wrong.

At the moment, you have implemented support for the Vertex AI Gemini API only for the Google Cloud Platform users.
This API requires specifying a projectId and location for authentication purposes, aligning with the cloud-based infrastructure.

However, I've found an alternative approach to interact with Gemini using Google AI Gemini API, that does not necessitate cloud dependency. This method involves generating an API key directly from Google AI Studio, which can then be seamlessly incorporated into requests either as a header or as a query parameter.

Should this be implemented as part of a separate starter, following the example of OpenAI and Azure OpenAI?

[prakash-aathi]

@DenisStrazhevich did you find any alternative solution ? is that spring Ai project supports gemini API key?