chore(provider): add Google AI Studio support

[tivris]

Add Google AI Studio as a first-class provider, giving users direct access to Gemini models using their GEMINI_API_KEY without routing through OpenRouter.

Changes

Provider registration (provider.json):

• New google_ai_studio entry using the native Google API endpoint (https://generativelanguage.googleapis.com/v1beta)
• Auth via GEMINI_API_KEY (standard API key flow)
• response_type: "Google" to use the native Google response format rather than the OpenAI-compatible shim
• Curated model list: Gemini 3.1 Pro Preview, 3.1 Flash Lite Preview, 3 Pro Preview, 3 Flash Preview, 2.5 Pro, 2.5 Flash, and 2.0 Flash — all with 1M context window, tool support, and parallel tool calls; reasoning enabled on all except 2.0 Flash

Domain registration (provider.rs):

• Added ProviderId::GOOGLE_AI_STUDIO constant
• Registered in all_built_in() list
• Added display name mapping ("google_ai_studio" → "GoogleAIStudio")
• Added FromStr parsing for the "google_ai_studio" string
• Tests for display name, built-in inclusion, and FromStr parsing

Testing

# Set your Gemini API key
export GEMINI_API_KEY=your_key_here

# Run with a Gemini model via Google AI Studio provider
forge --model google_ai_studio/gemini-2.5-pro

# Run domain tests
cargo insta test --accept -p forge_domain

[mcgamer48ft]

Bump

[amitksingh1490]

Suggested change

	"response_type": "OpenAI",
	"response_type": "Google",

[amitksingh1490]

use this url https://generativelanguage.googleapis.com/v1beta/models/gemini-3-flash-preview:generateContent
refer vertex_ai

[amitksingh1490]

We have optimised the google response type for google models it will give you better & faster results

[mslinn]

I could not find any documentation on how to configure Gemini 3.1 models in forge.yaml.

[tivris]

I could not find any documentation on how to configure Gemini 3.1 models in forge.yaml.

Gemini 3.1 models don't require any special forge.yaml configuration, they're used just like any other provider. Here's the quickest way:

1. Run forge provider login google_ai_studio
2. Enter your Gemini API key (get one from https://aistudio.google.com/app/apikey)
3. Once inside forge, you can switch models with :model gemini-3.1-pro-preview (or :config-model to persist it)

Alternatively, you can set it in your .forge.toml / forge.yaml:

[session]
provider_id = "google_ai_studio"
model_id = "gemini-3.1-pro-preview"

[mslinn]

$ forge provider login google_ai_studio
? Enter your GoogleAIStudio API key: AIza...ezWf5uE
● [12:11:25] GoogleAIStudio configured successfully
● [12:11:25] ERROR: Bearer token is required in API key field

[tivris]

Seems to be weird behaviour coming from a previously configured Bedrock provider. After a successful login, forge tries to fetch models from all configured providers, and if Bedrock has stale/invalid credentials, it throws that error.

You can fix it by logging out of Bedrock:

forge provider logout bedrock

Then retry forge provider login google_ai_studio.

[mslinn]

That worked, which is weird because I have never used bedrock. Looks like additional error handling would help users. Thanks!

[tivris]

No worries! Yeah that's definitely a bug in how forge handles multiple configured providers. I just submitted a fix for it in #2936. Thanks for identifying it, that helped track down the root cause.