Google Gemini Live introduces memory for past conversation recall

Google has added persistent memory to its Gemini Live conversational AI, enabling the system to remember details from earlier interactions such as dietary preferences or family dates. The feature initially rolls out in English for users in the United States and stores up to 10,000 tokens per session for a 24‑hour window before automatic deletion in line with Google’s privacy policy.

Technical benchmarks show the memory layer adds roughly 120 ms of latency for cached queries, with response times rising to 180 ms or higher under heavy load. The update also requires API‑key rotation for existing integrations, prompting enterprises to audit dependencies and adjust caching strategies. Developers are advised to plan for the 24‑hour data expiry and possible cold‑start penalties when building real‑time applications.

Google’s documentation notes that the memory capability closes a gap with rival voice assistants that already support contextual recall, and broader language support is expected to follow the initial US rollout.