5.15 The find_conflicts tool

Read-only diagnostic that surfaces memories which may contradict each other: clusters of active memories that are both semantically close and share at least one tag (so they describe the same labelled topic but might disagree). Same clustering and output shape as find_near_duplicates; the differences are the lower default threshold and the tag gate.

The server does not decide whether the memories truly conflict — that is a judgement for the calling LLM (or you). Resolve a conflict by confirming the right memory, or by storing the corrected fact with remember using its supersedes field (the old fact is now wrong) or downweights (it is merely outdated).

5.15.1 Inputs

threshold

Number, 0.5–1.0. Optional, default 0.80 (lower than find_near_duplicates, since a conflict is “same topic, possibly different answer”).

max_groups

Integer. Optional, default 50. Hard cap 500.

5.15.2 Output

Identical shape to find_near_duplicates (groups of ids / max_similarity / pair_count / memories, plus a summary).

5.15.3 Errors

INVALID_INPUT, DATABASE_ERROR.