google-gemini-api

Use the @google/genai SDK to integrate Gemini API by setting your GEMINI_API_KEY environment variable and issuing prompts with the Gemini model. The SDK supports text generation, multimodal inputs (images, PDFs, video, audio), function calling, and grounding via Google Search across Gemini 2.5 family models.

Yes, Gemini API supports both parallel and compositional function calling through @google/genai. This enables production agentic workflows where the model selects and executes multiple functions in sequence or parallel, integrated with code execution and streaming endpoints.

Gemini 2.5 models support up to 1,048,576 input tokens. Context caching requires explicit model version suffixes and is useful for large-scale data analysis and multi-turn chat. Streaming endpoints are available to handle responses incrementally across long contexts.

Grounding via Google Search is built into the @google/genai SDK integration. You can configure retrieval patterns and enable real-time web grounding to augment generated responses with current information, useful for production applications requiring factual accuracy.

Yes, Gemini 2.5 models include thinking mode support through the @google/genai SDK. This enables extended reasoning for complex problem-solving within agentic workflows, though it increases token consumption within the 1,048,576 token window.

Avoid deprecated Gemini SDKs and use only @google/genai for production integration. The current SDK provides proper model version suffix handling, token window awareness, and access to the full feature set including function calling, streaming, and grounding.