v2.6.7
Cut multi-turn token costs with Gemini's Interactions API
A new GeminiInteractions model class builds on Google's stateful Interactions API, so agents can talk to the interactions endpoint directly instead of Gemini's generateContent. Rather than resending the full conversation on every turn, it stores prior turns server-side and references them by ID, so only the new message goes over the wire. That pulls down token cost and latency through implicit caching, and it opens the door to background execution for long-running work like Deep Research. The Agent class tracks the interaction ID for you, so multi-turn conversations just work.
Action required: Install google-genai>=2.0 to use it. The Interactions API is experimental and may change in future versions.
See the Gemini Interactions docs for the full capability set.
