Auto-resolve pending conflicts with an LLM judge (NDJSON stream)
What memloom conflicts auto calls. Re-judges every pending conflict with context
the dedup classifier never had: when each side was recorded and, for memories
distilled from a Claude Code import, the transcript excerpt behind it. Only a
decisive verdict resolves anything (keep_new, keep_existing, or keep_both,
applied through the same resolveConflict path as the manual endpoint above, so it’s
just as revertable); unsure leaves the conflict pending. Streams NDJSON: one
{"type":"item"} event per examined conflict, then a final {"type":"done"} event
with the totals. Needs the LLM; unavailable in offline mode.
Body
The body is of type object.
Response
NDJSON progress stream ending with a done (or error) event.
The response is of type string.
"{\"type\":\"item\",\"conflictId\":\"...\",\"index\":1,\"total\":3,\"verdict\":\"keep_new\",\"reason\":\"the newer session shows the change\",\"content\":\"the deploy target is railway\"}\n{\"type\":\"done\",\"examined\":3,\"resolved\":2,\"keepNew\":1,\"keepExisting\":0,\"keepBoth\":1,\"unsure\":1}\n"