Google enables Gemini API to combine Search, Maps, and custom functions in a single request

Google expanded the Gemini API: now built-in tools like Google Search and Google Maps can be combined with custom functions in a single request. This removes some of the manual orchestration around agent scenarios and brings the API closer to a format where the model itself coordinates search, code calls, and transitions between steps.

What Changed

The update announced by Google in March 2026 adds more cohesive tool handling to the Gemini API. Previously, developers often had to organize search separately, lay out a route to maps or external data separately, and then manually stitch the responses together in the application. Now much of this logic can be assembled within a single call: the model gains access to Google's built-in services and simultaneously can invoke custom functions if the scenario requires actions beyond the standard set.

A practical walkthrough shows five demos that gradually increase task complexity. The logic starts with a basic tool combination, then transitions to multi-step agent chains, where context transfer between stages, identification of parallel calls, and correct dialog continuation after an external function response are critical. For developers, this is a significant shift: orchestration spreads less across the backend and moves more into the model session itself. This makes the chain's behavior noticeably more transparent for debugging and testing.

How the Call Works

The key idea is that Gemini can now decide in a single pass when it needs web search, when it needs geospatial data from Maps, and when it needs a custom application function. If a question requires multiple actions, the model can maintain overall context, align results from different tools, and continue the chain without a complete scenario restart. Special emphasis is placed on parallel tool IDs and context circulation: this helps avoid confusing tool responses and transfer necessary data to the next step.

• Google Search pulls fresh information on the query topic
• Google Maps adds addresses, geographic context, and place data
• Custom functions plug in the application's internal business logic
• Multi-step chains enable building scenarios from multiple sequential actions

This approach is convenient for scenarios where the answer cannot be obtained from a single source. For example, an assistant can first find fresh information through Search, then verify a location on the map through Maps, then call an internal booking, calculation, or availability check function. Previously, such a sequence often had to be broken down into multiple requests and the state between them manually maintained. For services with real-world actions, this significantly simplifies the architecture.

Why This Matters for Developers

The main benefit is less glue between the model and product logic. Instead of writing a separate orchestrator for each tool combination, a team can describe functions, give the model access to the needed services, and build more natural agent scenarios on top. This is especially useful for assistants that should not just answer with text, but actually perform tasks: search for data, select a location on a map, verify parameters, pass the result to an internal service, and only then formulate the final answer for the user.

Another advantage is more predictable scaling of complex chains. When built-in Google tools and a company's own functions are connected in one process, the number of intermediate layers where context is usually lost or call logic breaks decreases. For teams, this means faster prototypes and less boilerplate code around them. And for products, a chance to move faster from a chatbot with suggestions to an agent that actually knows how to deliver a task to completion.

What This Means

Google is moving the Gemini API toward full-featured agent interfaces, where the model doesn't just generate text but manages a set of tools within a single session. If this approach proves stable in real production scenarios, developers will be able to assemble helpful AI assistants with less manual orchestration and get them to market faster.