{"version":"2.1.305","anchor":"memory-extraction-gets-a-cheap-first-look-pass-that-can-sk","canonical_anchor":"memory-extraction-gets-a-cheap-first-look-pass-that-can-sk","heading":"Memory extraction gets a quick first look that can skip the full run","tier":"internal","area":"Memory","scope":"individual","heads_up":false,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.305\/e\/memory-extraction-gets-a-cheap-first-look-pass-that-can-sk","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.305","markdown":"### Memory extraction gets a quick first look that can skip the full run\n\nBefore extracting memories, a small model now checks whether a memory write is likely and can skip the full extraction pass\n\n**Unclear.** Not settled whether anything runs this quick check in this release.\n\n**What**\n\nClaude Code can extract memories (notes saved for later sessions) from recent exchanges in the background.\n\n- A cheap first pass now asks a small model whether the recent exchanges are likely to produce a memory write, answering `{\"write_likely\": true|false}`.\n\n- If it fails, the full pass still runs.\n\n- Its verdicts (yes, no, empty or error) are recorded in usage reporting, alongside a new entry about subagent work in remote sessions.\n\n- A new route skips its work when `memorySave` is set.\n\n**Why**\n\nBackground memory extraction can skip model calls when nothing is likely to be saved, which can change when memories get written.\n\n- Area: Memory\n- Tier: Under the hood\n- Useful: 3\/5\n- Signal: 3\/5\n- Scope: individual\n- Heads-up: no"}