# google-deepmind-interactions-api-gemini-agents-2025-12-11

## Veille

Interactions API Google DeepMind - unified interface for models and Gemini agents

## Titre Article

Interactions API: A unified foundation for models and agents

## Date

2025-12-11

## URL

https://blog.google/technology/developers/interactions-api/

## Keywords

API, Gemini, AI agents, Google DeepMind, MCP, server state, remote tools, Deep Research, ADK, A2A

## Authors

Ali Çevik, Philipp Schmid

## Ton

Profile: Official product communication, accessible technical register, launch-announcement perspective.
Style: Enthusiastic yet measured corporate tone, precise technical vocabulary without excessive jargon. The article positions the product as a natural evolution ("models becoming systems") rather than a revolution. Authority conferred by the Google DeepMind byline. Target audience: developers and architects working with generative AI APIs.

## Pense-betes

- New API in public beta via the Gemini API in Google AI Studio
- Server-side state management (optional) - reduces client complexity
- Native support for the MCP (Model Context Protocol) protocol for remote tools
- Background execution without a required persistent connection
- Gemini Deep Research available as a built-in agent via the API
- Complementary to generateContent (which remains the main path for standard production)
- Integration with Agent Development Kit (ADK) and the Agent2Agent (A2A) protocol
- Clean schema supporting interleaved messages and tool calls

## RésuméDe400mots

Google DeepMind announces the public beta launch of the Interactions API, accessible through the Gemini API in Google AI Studio. This new interface offers a unified entry point for interacting with models such as Gemini and AI agents.

The API addresses the limitations of existing approaches by offering several key capabilities. First, server-side state management becomes optional, significantly reducing client-side code complexity. Developers no longer need to maintain the full conversation history locally.

Second, the API provides interpretable data structures for managing complex agent histories. The schema natively supports interleaved messages and multiple tool calls, facilitating the construction of sophisticated workflows.

Third, background execution capabilities enable long-running inference without maintaining persistent connections. This architecture is particularly suited to the "extended thinking" tasks that characterize modern agents.

Fourth, native support for the MCP (Model Context Protocol) protocol enables the integration of remote tools in a standardized way. This interoperability facilitates the composition of agents using external services.

The announcement also includes the availability of Gemini Deep Research in preview, a built-in research agent accessible directly through the API. This agent illustrates the direction Google is taking: models are evolving into complete systems capable of autonomous reasoning.

Developers note that this release is only a beginning, with plans to expand built-in agents and support custom agent development. The API is positioned as complementary to the existing generateContent endpoint, which remains recommended for standard production workloads requiring direct responses.

Integration is already available via the Agent Development Kit (ADK) and the Agent2Agent (A2A) protocol, with broader ecosystem adoption expected soon. This announcement is part of the broader industry trend toward standardized interfaces for agents, joining other players' efforts around the MCP protocol.

## GrapheDeConnaissance

- Google DeepMind —publie→ Interactions API (TECHNOLOGIE, 0.98)
- Interactions API —permet→ interface unifiée pour modèles et agents Gemini (CONCEPT, 0.95)
- Interactions API —utilise→ MCP (TECHNOLOGIE, 0.93)
- Interactions API —permet→ gestion d'état côté serveur (CONCEPT, 0.92)
- Interactions API —permet→ exécution en arrière-plan (CONCEPT, 0.9)
- Gemini Deep Research —fait_partie_de→ Interactions API (TECHNOLOGIE, 0.93)
- Ali Çevik —publie→ annonce de l'Interactions API (DOCUMENT, 0.85)
- Philipp Schmid —publie→ annonce de l'Interactions API (DOCUMENT, 0.85)
- Agent Development Kit (ADK) —utilise→ Interactions API (TECHNOLOGIE, 0.9)
- Agent2Agent (A2A) —utilise→ Interactions API (TECHNOLOGIE, 0.9)
- Google DeepMind —affirme_que→ Interactions API est complémentaire de generateContent (AFFIRMATION, 0.88)

---
Canonical: https://www.thekb.eu/en/fiches/google-deepmind-interactions-api-gemini-agents-2025-12-11/