lex-llm-gemini

LegionIO LLM provider extension for Google Gemini.

This gem lives under Legion::Extensions::Llm::Gemini and depends on lex-llm >= 0.4.3 for shared provider-neutral routing, response normalization, fleet responder execution, and schema primitives. It does not require or call legion-llm at runtime; legion-llm owns adapter creation, routing, and registry writes after discovering loaded lex-llm-* provider gems.

Load it with require 'legion/extensions/llm/gemini'.

• Legion::Extensions::Llm::Provider registration as :gemini
• Gemini-native chat requests through POST /v1beta/{model=models/*}:generateContent
• Streaming chat support through POST /v1beta/{model=models/*}:streamGenerateContent?alt=sse
• Model discovery through GET /v1beta/models
• Embeddings through POST /v1beta/{model=models/*}:embedContent
• Normalized chat, streaming, vision, function calling, and embedding capability mapping from supportedGenerationMethods
• Best-effort llm.registry availability events published to AMQP when transport is loaded
• Shared fleet/default settings via Legion::Extensions::Llm.provider_settings

Legion::Extensions::Llm::Gemini.default_settings
# {
#   provider_family: :gemini,
#   instances: {
#     default: {
#       endpoint: "https://generativelanguage.googleapis.com/v1beta",
#       tier: :frontier,
#       transport: :http,
#       credentials: { api_key: "env://GEMINI_API_KEY" },
#       usage: { inference: true, embedding: true },
#       limits: { concurrency: 4 }
#     }
#   }
# }

Legion::Extensions::Llm.configure do |config|
  config.gemini_api_key = ENV.fetch("GEMINI_API_KEY")
  config.gemini_api_base = "https://generativelanguage.googleapis.com/v1beta"
  config.default_model = "gemini-2.0-flash"
  config.default_embedding_model = "gemini-embedding-001"
end

Provider instances can opt in to consuming Legion LLM fleet requests. The provider-owned fleet actor only starts when at least one discovered instance enables fleet.respond_to_requests, and execution delegates to Legion::Extensions::Llm::Fleet::ProviderResponder from lex-llm.

extensions:
  llm:
    gemini:
      instances:
        local:
          fleet:
            enabled: true
            respond_to_requests: true
            capabilities:
              - chat
              - stream_chat
              - embed

The Gemini entry point and provider use Legion::Logging::Helper:

• list_models logs model discovery activity and publishes availability through the shared lex-llm registry publisher.
• Fleet request execution is delegated to the shared responder helper so acking, response publishing, and responder failures stay inside the provider-neutral fleet boundary.
• Registry publishing is best-effort and must not block provider calls when transport is unavailable.

bundle install
bundle exec rspec --format json --out tmp/rspec_results.json --format progress --out tmp/rspec_progress.txt
bundle exec rubocop -A

Apache-2.0