Normalize Gemini vs Google

[lyrixx]

ATM, the code base use Gemini or Google in class name, namespace, or configuration key.

It could be confusing I guess.
Anyway it's not predictable.

I think It would be better to stick to only one name (maybe Gemini).

WDYT?

[chr-hertel]

Man that's chaotic - but I have a hard time to get google's terminology straight here.

In Google's AI Studio you can get an API Key, what they call GEMINI_API_KEY in their examples, and they also call it Gemini API. But on the other hand Gemini is only one model family of Google, and there is Gemma or text embedding models as well. And Vertex AI. Oof ... 😆

[chr-hertel]

Okay, maybe we should have two bridges:

• Vertex -> rather the Google Cloud product for hosting all kinds of models
• Gemini -> access to Gemini API with that subset of models

@valtzu, you've been playing around with Google/Gemini mostly, right? What makes sense from your point of view?

[valtzu]

I think it's the simplest for us if we keep the bridges separate like you @chr-hertel suggested – because they are completely separate in Google's docs too and even though Gemini is available via Vertex, the request schema seemed slightly different (unless it's just outdated docs, getting plenty of 404 there 🙂).

I would prefer to have both (Google & Gemini/Vertex) in the fully-qualified class name and even configuration key, however it does not seem the current convention, f.e. Bedrock does not mention Aws/Amazon in the name.

I see @OskarStark already pushed a commit to change the current Google to Gemini, it makes sense to me, at least as long as we don't have Vertex implemented.

[chr-hertel]

Meaning for the current bridge to be renamed from Google to Gemini 👍

[developkariyer]

Since Google does not provide an official client for PHP, following the Python path might be more compatible to what Google aims - with the assumption that they will give priority to updating their own Python libraries to follow API changes.

With this in mind, current status is, there are two distinct libraries, which have completely different usage methodology:

google.genai: Used with API keys. Not all models are available in this API.
vertexai: Provides more models. Used with oauth2 authentication and I guess requires some google tools installed on server, at least it is what I did, in order not to fight with the documentation and do whatever google says to make things work.

For this specific case, for full naming compatibility with Google, what I suggest is:

Google\Genai, GoogleGenai or Genai (initial priority to fully implement)
Google\VertexAi, Google\Vertex\Ai or GoogleVertexAi (reserved to be implemented later)

I am new to repository and after having some insight, hope to actively support coding part also...
Regards.

[chr-hertel]

So the difference for us is that we can support one contract easily in different bridges.

So we could have a GeminiApi bridge that supports the Gemini models and key based authentication, and a VertexAI bridge that has a different auth mechanism.

we could go like this:

Symfony\AI\Platform\Bridge\Google
|- Gemini <- model stuff
|- GeminiPlatformFactory <- API Key based factory for Gemini API
|- VertexAIPlatformFactory <- factory for Vertex AI

GoogleContract should become GeminiContract tho

[OskarStark]

So based on the conversation, Claude summarized it like this:

---

Proposed Plan for Standardizing Google/Gemini Naming

Current Situation Analysis

The codebase has inconsistent naming between "Google" (provider) and "Gemini" (product):

• Namespaces use Google: Symfony\AI\Platform\Bridge\Google
• Main class is named Gemini
• Environment variable uses GEMINI_API_KEY
• Documentation mixes both: google-gemini-server-tools.rst

Recommended Approach: Standardize on "Gemini"

I propose standardizing on "Gemini" for the following reasons:

1. It's the actual product name users interact with
2. Prepares for potential future Vertex AI bridge (as suggested in comments)
3. Aligns with the environment variable already using GEMINI_API_KEY
4. Follows the pattern of other bridges (OpenAI, Anthropic use product names)

Implementation Steps

1. Rename namespaces and directories:

   • /src/platform/src/Bridge/Google/ → /src/platform/src/Bridge/Gemini/
   • Update namespace from Symfony\AI\Platform\Bridge\Google to Symfony\AI\Platform\Bridge\Gemini

2. Update class names:

   • GoogleContract → GeminiContract
   • Keep Gemini class name (already correct)
   • Update factory references

3. Update service definitions:

   • ai.platform.contract.google → ai.platform.contract.gemini

4. Update documentation:

   • Rename google-gemini-server-tools.rst → gemini-server-tools.rst
   • Update all references consistently

5. Update examples:

   • /examples/google/ → /examples/gemini/

Future Considerations

• This change prepares for a potential separate Vertex AI bridge in the future
• Vertex AI would get its own namespace: Symfony\AI\Platform\Bridge\VertexAI
• Clear separation between direct Gemini API and Google Cloud Vertex AI

---

Do you agree?

[chr-hertel]

From my point of view, yes. @lyrixx do you See your concerns resolved with that?

[lyrixx]

Perfect 👍🏼