AI Dreams: A Long-Term Memory Pattern for AI Agents by Jason Davenport

🛠️ AI Dreams: A Long-Term Memory Pattern for AI Agents

Jason Davenport published the 'electric-sheep' demo project based on Google ADK 2.0. The concept is offline memory consolidation, analogous to REM sleep: BigQuery ML aggregates sessions, Cloud Spanner stores embeddings, and vector search takes <10ms without LLM token costs.

🌍 The offline consolidation pattern on BigQuery ML + Spanner vector search shows that agent memory doesn't require specialized vector DBs or additional dependencies.

👤 A ready-made project with UI, Terraform, and instructions — you can deploy it on your own GCP. Applicable to any agentic framework, not just Google ADK.

Source 1: https://davenporter.substack.com/p/give-an-agent-access-to-memories Source 2: https://github.com/davenportjw/electric-sheep