After 400+ agents, an engineer maps 15 AI agent memory patterns from types to control
MaryamMiradi · x · 2026-10-20
Engineer Maryam Miradi, who has built 400+ AI agents, curates 15 memory patterns in three groups, with open-source GitHub links: Memory Types (working, episodic, semantic, procedural, long-term), Memory Mechanisms (summarization, retrieval, consolidation, compression, graph memory), and Memory Control (decay, forgetting, conflict resolution, provenance, isolation). Together they form one memory architecture: what to remember, how to process it, and what to trust, forget, or isolate.
More from coding & agent
- Coinbase launches perps trading for AI agents: Claude, ChatGPT and Grok can now go long US indices — MurrLincoln · 2026-10-20
- Google Cloud rethinks Kubernetes for agents: suspend/resume as core primitive — rseroter · 2026-10-20
- This MCP server is under 700 bytes: one time tool, zero dependencies, instant startup — kaphwor · 2026-10-20
- Anthropic Dev Shares the Prompts Behind His Opus 5.5 Code-Rendered Films — dfeinition · 2026-10-20
- Meta's ProPerAgent lifts memory-driven proactive assistants from 15% to 38% — dair_ai · 2026-10-20
- Claude Motion animations now send straight to Descript for full video production — descript · 2026-10-20