Interactions API

Overview

The Interactions API is Google DeepMind's Gemini API interface for orchestrating models and agents through standard REST patterns plus Server-Sent Events (SSE), rather than relying on more bespoke RPC-style behavior. It was introduced as a unified interface for Gemini models and agents, then later moved into general availability as the default interface for combining model reasoning, tool use, and agent workflows.

For AI Product Managers, the Interactions API matters because it represents a product design choice with strategic implications: simplifying developer experience by adopting familiar web standards. That makes integration easier, error handling more understandable, and multi-agent or tool-using workflows more accessible to product teams shipping real applications. It also signals how platform providers are packaging agent capabilities into more standard, production-friendly APIs that can be documented via OpenAPI, tested with curl, and integrated into existing backend systems with less custom infrastructure.

Key Developments

• 2026-01-03: Google DeepMind announced the Interactions API in beta. Phil Schmid described it as a unified interface for Gemini models and agents that would power agentic features in 2026.
• 2026-05-05: Logan Kilpatrick rolled out major error-message improvements for the Interactions API, making responses more readable for both humans and agents.
• 2026-05-05: Philipp Schmid highlighted quality-of-life upgrades to validation errors, including exact field names, bad values, supported enum options, expected vs. actual formats, and precise field paths such as `input[0].name`.
• 2026-05-23: At Google I/O, Philipp Schmid demoed building an AI agent with a secure hosted Linux sandbox in a single API call using Gemini Managed Agents and the Interactions API to execute code and manage memory.
• 2026-05-23: Additional clarification noted that each sandbox session gets its own persistent container ID, which can be shared across agents, with a tradeoff of slower cold starts in exchange for persistence.
• 2026-06-23: Logan Kilpatrick announced the Interactions API had reached GA and become the new default interface for orchestrating models and agents together.
• 2026-07-29: Philipp Schmid described the GA launch as replacing RPC quirks with standard REST + SSE for a more consistent developer experience, with quickstart curl examples and an OpenAPI spec published in the Gemini API docs.

Relevance to AI PMs

• Evaluate platform maturity through interface design. PMs can use the Interactions API as a benchmark for whether an AI platform is moving from experimental tooling toward production-ready developer experience. Standard REST, SSE streaming, and OpenAPI support reduce onboarding friction for internal teams and external developers.
• Improve integration planning for agent features. If your roadmap includes tool use, code execution, memory, or multi-step orchestration, the Interactions API shows how those capabilities can be exposed through a single unified surface instead of fragmented endpoints or custom orchestration layers.
• Prioritize better observability and debugging in AI products. The newsletter mentions around error-message upgrades are especially relevant for PMs managing developer platforms. Clear field-level validation, supported enum hints, and expected-vs-actual formatting reduce support burden and speed time-to-resolution for customers.

Related

• Google DeepMind: The organization behind the Interactions API and the broader Gemini platform strategy.
• Gemini / Gemini API: The Interactions API is part of the Gemini API stack and serves as a core interface for model and agent orchestration.
• Phil Schmid / Philipp Schmid: Frequently associated with announcing and demonstrating the API, including beta, GA, and Google I/O demos.
• Logan Kilpatrick: Associated with the GA rollout and improvements to error handling and developer experience.
• Gemini Managed Agents: Closely connected through demos showing secure hosted sandboxes, code execution, and memory managed via the Interactions API.
• Agents: The API is positioned as a default interface for building and orchestrating agentic workflows.
• OpenAPI: Important because the published spec makes the API easier to inspect, generate clients for, and integrate into standard engineering workflows.
• OpenAI: Relevant as a neighboring ecosystem player; comparisons may be useful for PMs evaluating how different vendors expose agent orchestration patterns and developer abstractions.