Google Deepmind upgrades Gemini API with multi-tool chaining and context circulation

Editor's take

Google Deepmind's Gemini API now allows developers to orchestrate sequences of tool calls and integrate real-time location data from Google Maps. This enhancement moves beyond single-function calls, enabling more complex workflows where the output of one tool can directly inform the input of another, and vice-versa, creating a dynamic feedback loop.

This development matters because it addresses a key limitation in current LLM development: the ability to perform multi-step reasoning and access external, dynamic information. For businesses building AI agents that require real-world context, such as travel planning or logistics, this offers a more robust foundation than previous models that were largely confined to their training data or required cumbersome manual orchestration.

Future developments to monitor include the performance and latency of these chained calls, particularly with multiple tool integrations. The ability for Gemini to efficiently manage context circulation, avoiding information decay across numerous steps, will be crucial for its adoption in mission-critical applications. Additionally, observing how other LLM providers, like OpenAI with its Assistants API, respond to this multi-tool capability will indicate the direction of agent development.