Google adds context circulation, tool combos, and Maps grounding to the Gemini API

Google on Mar 17, 2026 announced a set of Gemini API updates aimed at teams building agentic workflows on top of Gemini 3. The release focuses on reducing orchestration friction when a model needs to move between built-in tools and application-specific functions during a single task.

The most visible change is that developers can now pass built-in tools and custom function declarations in the same request. Google says this lets Gemini switch from built-in capabilities such as Google Search or Google Maps to a developer backend without forcing the application to split the workflow across separate requests. For production systems, that can remove glue code, reduce latency, and simplify the way retrieval and action are connected.

Key updates

• Built-in tools and custom functions can now be combined in the same interaction.
• Context circulation keeps tool calls and responses available across later reasoning steps.
• Each tool call gets a unique identifier to improve asynchronous execution and debugging.
• Grounding with Google Maps now works across the Gemini 3 family for location-aware responses.

Google also introduced context circulation for built-in tools. That means the outputs of earlier tool calls stay in the model context so later steps can reuse them without requiring the client application to manually restate every intermediate result. The company added unique call identifiers for each tool call as well, which should make asynchronous and parallel tool execution easier to debug and reconcile.

A second important change is the expansion of Grounding with Google Maps to the Gemini 3 family. Google positions this as a way to give models access to up-to-date spatial context, place details, commute times, and local business information. For developers building travel, commerce, logistics, field service, or local discovery products, the update broadens the range of location-aware tasks Gemini can support.

Google said the same capabilities are also supported in the generateContent API, but recommended the newer Interactions API for server-side state management and unified reasoning traces. Taken together, the announcement is less about a new model release and more about making Gemini easier to operate as an execution layer for real-world multi-tool agents.

Source: Google