Google DeepMind has introduced the Interactions API, a unified interface for interacting with Gemini models and agents like Gemini Deep Research.
What is the Interactions API?
A single RESTful endpoint that simplifies building advanced agentic applications by handling:
- •Server-side state management - Reduces client complexity and context errors
- •Background execution - Long-running inference without maintaining connections
- •MCP tool support - Direct calls to Model Context Protocol servers
- •Cache optimization - Offload history management to reduce costs
How It Works
Developers can interact with:
- •Models - Using the
modelparameter (e.g., Gemini 3 Pro) - •Agents - Using the
agentparameter (e.g., deep-research-pro-preview)
python# Example: Using with Gemini Deep Research agent
response = client.interactions.create(
agent="deep-research-pro-preview",
messages=[{"role": "user", "content": "Research quantum computing advances"}]
)
Key Features
Feature | Benefit |
|---|---|
| Single endpoint | Unified access to models and agents |
| Server-side state | No client-side history management |
| Background tasks | Fire-and-forget long research tasks |
| MCP integration | Connect external tools seamlessly |
4 items
Availability
- •Status: Public beta via Gemini API
- •Access: Google AI Studio with API key
- •Coming soon: Vertex AI support
Note: Still in beta, subject to breaking changes. generateContent remains the primary production path.
Source: Google Blog