feat: Add provide experimental support for Google Gemini #67

[shanehuntley]

Basic provider support for Google Gemini models.

Passes local and integration tests. Requires GOOGLE_API_KEY to be set for use but works with Gemini API Free Tier.

Example Goose configuration:

default:
provider: google
processor: gemini-1.5-pro
accelerator: gemini-1.5-flash
moderator: passive
toolkits:

• name: developer
  requires: {}

Fixes #67

[codefromthecrypt]

cool start. Is there any way you can add VCR tests such as exist in openai? If not, I can I guess sign up for access and fill them later.

Also, did you manage to try vision? I didn't look up if the free tier includes a vision capable model. If so, give a whirl in test_integration.py!

Finally, do we need an entry in pyproject.toml under here?

[project.entry-points."exchange.provider"]

[shanehuntley]

I haven't looked at VCR tests or the vision aspects yet. Both gemini flash and pro do support vision.

[codefromthecrypt]

Both gemini flash and pro do support vision.

quick try is to add a tuple to test_integration_vision.py and see if it works?

[baxen]

LGTM!

Agree we should pursue VCR and vision tests but happy for those to be in another PR, since this looks to be working for the core use cases