feat: add Google Gemini provider support

[do420]

Summary

Adds first-class Google Gemini support as an AI provider using Gemini's OpenAI-compatible endpoint.

Changes

src/providers.ts

• Added gemini entry to PROVIDERS map using the OpenAI-compatible base URL https://generativelanguage.googleapis.com/v1beta/openai
• Added gemini to PROVIDER_ENV_VARS so GEMINI_API_KEY env var is automatically detected

src/openclaw-credentials.ts

• Added Gemini URL pattern to inferProviderFromBaseUrl() so the provider is correctly identified from the base URL
• Fixed
  esolveApiConfig() to propagate localProvider in both return paths (it was being read to find the right env key but then dropped from the returned object)

src/ai-brain.ts

• Added gemini to AIBrain.BASE_URLS fallback map so the correct endpoint is used when �aseUrl isn't explicitly set

src/index.ts

• Fixed �isionModel/model falling back to the pipeline config's layer3/layer2 model names when
  esolveApiConfig() returns no model - prevents an empty string being sent to the API causing silent failures in the vision loop

Configuration

Set these env vars (or add to .env):

.env GEMINI_API_KEY=your_key_here AI_BASE_URL=https://generativelanguage.googleapis.com/v1beta/openai AI_TEXT_MODEL=gemini-2.0-flash AI_VISION_MODEL=gemini-2.0-flash

Or start with explicit flags:
clawdcursor start --provider gemini --api-key your_key_here

Testing

Tested end-to-end on Windows with gemini-2.0-flash:

• YouTube search task (navigate + type + click) ?
• Gmail compose + send flow ?

[AmrDab]

hey @do420 , these problems have been addressed in the V0.7.0 update on branch, check it out. also what OS are you operating on?

[do420]

hey @do420 , these problems have been addressed in the V0.7.0 update on branch, check it out. also what OS are you operating on?

Thanks! Any plans on merging v0.7.0 to main? I am operating on Windows.