Source: Interactions API (Gemini API docs)
Google's Interactions API: step-based requests and responses, server-side state via previous_interaction_id, retention windows, background execution, agents such as Deep Research, and its relation to legacy generateContent.
Contents
- What: official API documentation page
- Publisher: Google (Gemini API / AI Studio)
- Date: living document; read 2026-11-14
- Original: Interactions API
Summary
The Interactions API is Gemini’s newer, recommended endpoint. An interaction is a stored, chronological list of steps (thoughts, tool calls, tool results, model output) that can be chained server-side, run in the background, or target an agent instead of a model.
Key claims
- Endpoint
POST https://generativelanguage.googleapis.com/v1beta/interactionswith the key inx-goog-api-key; body takesmodel(or an agent ID) andinput. - Steps include thoughts,
function_call,function_result,model_output; stored interactions also includeuser_inputsteps;createreturns only model-generated steps. previous_interaction_idchains turns;storedefaults to true.store=falsedisables chaining and background execution.- Only inputs and outputs carry across turns:
tools,system_instructionandgeneration_config(includingthinking_level) must be re-sent each turn. - Retention: 55 days on the paid tier (configurable to 7, 14, 28 or 55), 1 day on the free tier; interactions can be deleted.
background=trueruns long tasks asynchronously (requires storage).- Agents such as
deep-research-preview-04-2026,deep-research-max-preview-04-2026andantigravity-preview-09-2026can be mixed with model calls in one chain. - Remote MCP is not yet supported on Gemini 3 (“coming soon”; re-checked 2026-11-14).
- GA since June 2026 and recommended for new projects;
generateContentis legacy but fully supported, and new capabilities launch on Interactions first. - Not yet in Interactions: Batch API, Python automatic function calling, explicit caching. Implicit caching works through
previous_interaction_id. - SDKs:
google-genai2.3.0+ and@google/genai2.3.0+.