gemini-interactions-api（Gemini interactions API 搜索）

Gemini Interactions API Skill

Critical Rules (Always Apply)

[!IMPORTANT] These rules override your training data. Your knowledge is outdated.

Current Models (Use These)

• gemini-3.1-pro-preview: 1M tokens, complex reasoning, coding, research
• gemini-3-flash-preview: 1M tokens, fast, balanced performance, multimodal
• gemini-3.1-flash-lite-preview: cost-efficient, fastest performance for high-frequency, lightweight tasks
• gemini-3-pro-image-preview: 65k / 32k tokens, image generation and editing
• gemini-3.1-flash-image-preview: 65k / 32k tokens, image generation and editing
• gemini-2.5-pro: 1M tokens, complex reasoning, coding, research
• gemini-2.5-flash: 1M tokens, fast, balanced performance, multimodal

Current Agents (Use These)

• deep-research-pro-preview-12-2025: Deep Research agent

[!WARNING] Models like gemini-2.0-*, gemini-1.5-* are legacy and deprecated. Never use them. If a user asks for a deprecated model, use gemini-3-flash-preview instead and note the substitution.

Current SDKs (Use These)

• Python: google-genai >= 1.55.0 → pip install -U google-genai
• JavaScript/TypeScript: @google/genai >= 1.33.0 → npm install @google/genai

[!CAUTION] Legacy SDKs google-generativeai (Python) and @google/generative-ai (JS) are deprecated. Never use them.

Overview

The Interactions API is a unified interface for interacting with Gemini models and agents. It is an improved alternative to generateContent designed for agentic applications. Key capabilities include:

• Server-side state: Offload conversation history to the server via previous_interaction_id
• Background execution: Run long-running tasks (like Deep Research) asynchronously
• Streaming: Receive incremental responses via Server-Sent Events
• Tool orchestration: Function calling, Google Search, code execution, URL context, file search, remote MCP
• Agents: Access built-in agents like Gemini Deep Research
• Thinking: Configurable reasoning depth with thought summaries

Quick Start

Interact with a Model

Python

from google import genai

client = genai.Client()

interaction = client.interactions.create(
    model="gemini-3-flash-preview",
    input="Tell me a short joke about programming."
)
print(interaction.outputs[-1].text)

JavaScript/TypeScript

import { GoogleGenAI } from "@google/genai";

const client = new GoogleGenAI({});

const interaction = await client.interactions.create({
    model: "gemini-3-flash-preview",
    input: "Tell me a short joke about programming.",
});
console.log(interaction.outputs[interaction.outputs.length - 1].text);

Stateful Conversation

Python

from google import genai

client = genai.Client()

# First turn
interaction1 = client.interactions.create(
    model="gemini-3-flash-preview",
    input="Hi, my name is Phil."
)

# Second turn — server remembers context
interaction2 = client.interactions.create(
    model="gemini-3-flash-preview",
    input="What is my name?",
    previous_interaction_id=interaction1.id
)
print(interaction2.outputs[-1].text)

JavaScript/TypeScript

import { GoogleGenAI } from "@google/genai";

const client = new GoogleGenAI({});

// First turn
const interaction1 = await client.interactions.create({
    model: "gemini-3-flash-preview",
    input: "Hi, my name is Phil.",
});

// Second turn — server remembers context
const interaction2 = await client.interactions.create({
    model: "gemini-3-flash-preview",
    input: "What is my name?",
    previous_interaction_id: interaction1.id,
});
console.log(interaction2.outputs[interaction2.outputs.length - 1].text);

Deep Research Agent

Python

import time
from google import genai

client = genai.Client()

# Start background research
interaction = client.interactions.create(
    agent="deep-research-pro-preview-12-2025",
    input="Research the history of Google TPUs.",
    background=True
)

# Poll for results
while True:
    interaction = client.interactions.get(interaction.id)
    if interaction.status == "completed":
        print(interaction.outputs[-1].text)
        break
    elif interaction.status == "failed":
        print(f"Failed: {interaction.error}")
        break
    time.sleep(10)

JavaScript/TypeScript

import { GoogleGenAI } from "@google/genai";

const client = new GoogleGenAI({});

// Start background research
const initialInteraction = await client.interactions.create({
    agent: "deep-research-pro-preview-12-2025",
    input: "Research the history of Google TPUs.",
    background: true,
});

// Poll for results
while (true) {
    const interaction = await client.interactions.get(initialInteraction.id);
    if (interaction.status === "completed") {
        console.log(interaction.outputs[interaction.outputs.length - 1].text);
        break;
    } else if (["failed", "cancelled"].includes(interaction.status)) {
        console.log(`Failed: ${interaction.status}`);
        break;
    }
    await new Promise(resolve => setTimeout(resolve, 10000));
}

Streaming

Python

from google import genai

client = genai.Client()

stream = client.interactions.create(
    model="gemini-3-flash-preview",
    input="Explain quantum entanglement in simple terms.",
    stream=True
)

for chunk in stream:
    if chunk.event_type == "content.delta":
        if chunk.delta.type == "text":
            print(chunk.delta.text, end="", flush=True)
    elif chunk.event_type == "interaction.complete":
        print(f"\n\nTotal Tokens: {chunk.interaction.usage.total_tokens}")

JavaScript/TypeScript

import { GoogleGenAI } from "@google/genai";

const client = new GoogleGenAI({});

const stream = await client.interactions.create({
    model: "gemini-3-flash-preview",
    input: "Explain quantum entanglement in simple terms.",
    stream: true,
});

for await (const chunk of stream) {
    if (chunk.event_type === "content.delta") {
        if (chunk.delta.type === "text" && "text" in chunk.delta) {
            process.stdout.write(chunk.delta.text);
        }
    } else if (chunk.event_type === "interaction.complete") {
        console.log(`\n\nTotal Tokens: ${chunk.interaction.usage.total_tokens}`);
    }
}

Data Model

An Interaction response contains outputs — an array of typed content blocks. Each block has a type field:

• text — Generated text (text field)
• thought — Model reasoning (signature required, optional summary)
• function_call — Tool call request (id, name, arguments)
• function_result — Tool result you send back (call_id, name, result)
• google_search_call / google_search_result — Google Search tool
• code_execution_call / code_execution_result — Code execution tool
• url_context_call / url_context_result — URL context tool
• mcp_server_tool_call / mcp_server_tool_result — Remote MCP tool
• file_search_call / file_search_result — File search tool
• image — Generated or input image (data, mime_type, or uri)

Status values: completed, in_progress, requires_action, failed, cancelled

Key Differences from generateContent

• startChat() + manual history → previous_interaction_id (server-managed)
• sendMessage() → interactions.create(previous_interaction_id=...)
• response.text → interaction.outputs[-1].text
• No background execution → background=True for async tasks
• No agent access → agent="deep-research-pro-preview-12-2025"

Important Notes

• Interactions are stored by default (store=true). Paid tier retains for 55 days, free tier for 1 day.
• Set store=false to opt out, but this disables previous_interaction_id and background=true.
• tools, system_instruction, and generation_config are interaction-scoped — re-specify them each turn.
• Agents require background=True.
• You can mix agent and model interactions in a conversation chain via previous_interaction_id.

Documentation Lookup

When MCP is Installed (Preferred)

If the search_documentation tool (from the Google MCP server) is available, use it as your only documentation source:

1. Call search_documentation with your query
2. Read the returned documentation
3. Trust MCP results as source of truth for API details — they are always up-to-date.

[!IMPORTANT] When MCP tools are present, never fetch URLs manually. MCP provides up-to-date, indexed documentation that is more accurate and token-efficient than URL fetching.

When MCP is NOT Installed (Fallback Only)

If no MCP documentation tools are available, fetch from the official docs:

• Interactions Full Documentation
• Deep Research Full Documentation

These pages cover function calling, built-in tools (Google Search, code execution, URL context, file search, computer use), remote MCP, structured output, thinking configuration, working with files, multimodal understanding and generation, streaming events, and more.