{"slug":"long-term-memory","label":"Long-Term Memory","item_count":3,"day_count":3,"source_count":3,"first_seen":"2026-09-18T15:55:29+00:00","last_updated":"2026-10-05T03:31:51+00:00","generated_at":"2026-10-06T05:05:15.739864+00:00","sources":["arxiv_cs_ai","arxiv_cs_lg","hackernews_ai"],"days":[{"date":"2026-09-18","items":[{"title":"AutoViewMem: Self-Configuring Orthogonal Views for Conversational Long-Term Memory","url":"http://arxiv.org/abs/2609.21940v1","source":"arxiv_cs_ai","type":"paper","summary_1line":"Long-term memory is essential for large language model (LLM) agents to maintain consistency and personalization over extended interactions. Existing memory systems typically rely on fixed granularities or static schem...","why_it_matters":"Matches feed focus: agent, eval.","sid":"eddc39b344540bb2","published":"2026-09-18T15:55:29+00:00","editor_note":"Paper proposing self-configuring memory views for conversational agents."}]},{"date":"2026-09-22","items":[{"title":"Beyond Context Windows: Evaluating Long-Term Memory for AI Agents","url":"https://twitter.com/AgentMemoryL/status/2101312784688726331","source":"hackernews_ai","type":"news","summary_1line":"Beyond Context Windows: Evaluating Long-Term Memory for AI Agents","why_it_matters":"Matches feed focus: agent, eval.","sid":"e1b58c64b7362aaf","published":"2026-09-22T06:40:56+00:00","editor_note":"Frames evaluation of long-term agent memory, a gap retrieval-only tests leave open."}]},{"date":"2026-10-05","items":[{"title":"PACMI: Provenance-Aware Cascading Memory Invalidation for Long-Term LLM Agents","url":"http://arxiv.org/abs/2610.05732v1","source":"arxiv_cs_lg","type":"paper","summary_1line":"LLM agents rely on long-term memory to retain and reuse information when performing tasks over long horizons. Existing methods provide limited support for handling memories that become outdated as new observations or...","why_it_matters":"Matches feed focus: agent, evaluation.","sid":"7866a5b7fe98e94b","published":"2026-10-05T03:31:51+00:00","editor_note":"Adds provenance-aware invalidation for memories that go stale."}]}],"editorial":{"tldr":"Agent memory research started with how to store conversational memory: AutoViewMem proposed self-configuring views instead of fixed granularities. A benchmark post then asked how to evaluate long-term memory beyond the context window.","stale":false,"whats_new":"PACMI (Oct 5) tackles stale memory: it invalidates dependent memories when an upstream fact is outdated, tracked through provenance.","why_it_matters":"Agents that persist memory across sessions need an update and invalidation policy, not only a retrieval layer; stale memories are a reliability risk that retrieval benchmarks do not catch.","take_for_builders":"If your agent persists memory across sessions, record where each memory came from and test that correcting one fact also retires memories derived from it.","status":{"state":"Developing","tone":"rising","changed":"2026-10-05","detail":"Research focus is moving from how to store memory to how to keep it correct and measure it."},"beats":[{"kicker":"STORAGE","tone":"launch","headline":"AutoViewMem configures memory views per conversation","summary":"Replaces fixed granularities and static schemas in conversational memory.","sids":["eddc39b344540bb2"]},{"kicker":"EVALUATION","tone":"rising","headline":"A benchmark post asks how to evaluate memory beyond context windows","sids":["e1b58c64b7362aaf"]},{"kicker":"NOW","tone":"now","headline":"PACMI cascades invalidation through provenance-linked memories","summary":"Targets memories that become outdated as new observations arrive.","sids":["7866a5b7fe98e94b"]}],"open_questions":["Does PACMI's invalidation cascade hold up on an independent long-horizon agent benchmark?","Will the long-term memory benchmark be extended to score stale-memory handling?"],"generated_at":"2026-10-06T05:20:00+00:00"}}