recall

Recall from Unified Memory

Surface relevant memories — raw conversation/doc chunks AND crystallized vault lessons in one ranking — lay your planned actions against them, and report, out loud, whether memory changed the plan.

Overview

Memory has two layers retrieved in ONE call: raw chunks (every past conversation and doc, embedded mechanically) and vault notes (lessons crystallized from them). Recall's jobs, in order:

1. Make your plan visible before retrieving anything — an unstated plan cannot be tested against memory.
2. Sweep, then run ONE unified engram query. Items tagged kind: chunk are raw fragments; kind: fact/feedback/runbook are crystallized lessons. They compete in the same top-N.
3. Crystallize — when several near-match chunks evidence the same principle and no note states it yet, write the vault note now.
4. Explore-sample and ride-along — the binary samples additional notes from vocab-term centroids near the query (budget sized to the matched-note count) and inserts superseded-note supersessors at the next rank; the agent judges the surfaced candidates, never links.
5. Synthesize impact on the plan — confirm / adjust / contradict / silent, per planned action.
6. Re-enter for emergent recommendations — a recommendation conceived mid-work gets its own lever-keyed query and a Re-entry: line directly above it before it ships (Step 3.5).

The binary resolves the vault and chunk index automatically ($XDG_DATA_HOME/engram/...; ENGRAM_VAULT_PATH / ENGRAM_CHUNKS_DIR override). Do not pass --vault or --chunks-dir.

Modes — glance vs deep (the depth dial)

Recall runs in one of two modes, selected by the caller (the mode word is the skill argument; absent → deep):

• deep (default). The full procedure below — all 10 phrases and the write side (Steps 2.5C, Step 4). It both applies memory to this decision and grows the vault (crystallizes, persists synthesis). Use it when the decision is weighty or irreversible, when you want recall to also learn, or when in doubt.
• glance (opt-in, cheap — for firing often). A pass that is read-only with respect to vault knowledge (Step 2.7 activate still bumps the used-notes recency metadata and pulls a used from_parent note down as a pending copy — both kept, neither a knowledge write; glance leaves that copy pending, and curating it is a separate curate-skill step after the user's request, not part of glance). Run Steps 0–3.5 with ~3 phrases (not 10) and keep the read side — Step 2.5A (read candidates), Step 2.5B (apply the recency weight), Step 2.7 (activate used notes), the Step 3 synthesis, and Step 3.5 (the re-entry query, when triggered) — but skip the write side: Step 2.5C (coverage amend/learn), Step 4 (synthesis-persist). Glance applies memory to this decision; it does not grow the vault's knowledge.