A real LoCoMo conversation, walked through the T-Mem write pipeline stage by stage — segmented scenes, extracted facts, and all four trigger families. Then a few real questions answered from the memory graph.
Pre-computed artifacts. This explorer renders stored outputs from the paper's runs (memory-construction LLM: GPT-4.1-mini; encoder: bge-m3). Building memory live needs LLM API keys — the Run locally tab shows how to reproduce it end-to-end.