Gemini API: Multi-Tool Chaining, Context Circulation, and Maps Grounding for Gemini 3

Overview

On March 18, 2026, Google released a significant set of tooling improvements to the Gemini API, enabling more sophisticated agentic workflows for developers building with Gemini 3 models. The update touches three interconnected areas: multi-tool combination, context circulation, and expanded Google Maps grounding support.

Multi-Tool Combination

Previously, developers who needed Gemini to use both a built-in tool (such as Google Search) and a custom function had to structure these as separate sequential requests, manually passing outputs between them. The March 18 update eliminates this friction entirely.

Developers can now pass both google_search (or other built-in tools) and function_declarations in the same tools parameter of a single API call. Gemini automatically decides when to invoke each tool and in what order, pivoting between fetching public data via Google Search and calling a developer's own backend APIs without requiring separate orchestration steps. This reduces end-to-end latency and significantly simplifies agent architecture.

To enable this capability, developers must set include_server_side_tool_invocations=True in their GenerateContentConfig. Google recommends pairing this with the new Interactions API, which handles server-side state management and provides unified reasoning traces across tool calls.

Context Circulation

A key challenge in multi-step tool workflows has been that built-in tool outputs were not automatically visible to subsequent tool calls — developers had to extract and re-inject that context manually. Context circulation solves this.

With context circulation enabled, every tool call and its full response is preserved in the model's running context window, turn after turn. A subsequent custom function call can directly reference what Google Search returned two steps earlier, without any manual plumbing. Google illustrates the pattern with a concrete example: Gemini can call a weather built-in tool, and the returned weather data automatically feeds into a custom venue-booking function that needs to know current conditions before placing a reservation.

Alongside context circulation, Google introduced unique call identifiers (id) for every tool invocation. These IDs allow developers to precisely map each model-requested tool call to the corresponding client response — which is especially critical when handling parallel function calling where multiple tools execute simultaneously.

Grounding with Google Maps for Gemini 3

Grounding with Google Maps is now fully supported across the Gemini 3 model family, including Gemini 3.1 Pro Preview, Gemini 3.1 Flash-Lite Preview, Gemini 3 Flash Preview, and the Gemini 2.5 series.

The feature connects Gemini's reasoning to Google's database of over 250 million places worldwide, enabling a new class of geospatially-aware AI applications. When a user query has geographic context, the model automatically detects this and draws on Google Maps data to provide grounded, accurate responses — covering local business information, commute times, place details, and real-time availability.

Developers enable the feature by including a googleMaps object in the tools parameter. Responses include groundingMetadata with source URIs and place IDs, plus an optional googleMapsWidgetContextToken that can be used to render an interactive Places widget in the application UI — giving end users a visually rich map experience alongside generated content.

Pricing: $25 per 1,000 grounded prompts, with a free tier of 500 daily requests. Requests are only billed when at least one Maps source appears in the response.

What This Means for Developers

Together, these three capabilities represent a meaningful step toward making Gemini a first-class platform for agentic applications. The combination of multi-tool support and context circulation means developers can build workflows where Gemini autonomously orchestrates a chain of heterogeneous tools — built-in and custom — with full memory of what each step returned. The Maps grounding extension gives location-aware applications a powerful grounding source without requiring developers to manage their own Places API integration separately.