Deleted transcripts are ingested by dreaming because the heartbeat filter is incomplete, leaking heartbeat noise into the dream corpus. Users who delete sessions to remove bad data find that the deleted content is still being processed by the memory system — the deletion does not
Deleted transcripts are ingested by dreaming because the heartbeat filter is incomplete, leaking heartbeat noise into the dream corpus. Users who delete sessions to remove bad data find that the deleted content is still being processed by the memory system — the deletion does not
Deleted transcripts are ingested by dreaming because the heartbeat filter is incomplete, leaking heartbeat noise into the dream corpus. Users who delete sessions to remove bad data find that the deleted content is still being processed by the memory system — the deletion does not actually prevent ingestion. The dream/memory system operates on garbage data that users explicitly tried to remove.
The memory_search tool returns 0 results for corpus=sessions, while the CLI equivalent works fine. The same corpus, same search, produces results from the command line but returns nothing when called via the tool interface. Users relying on memory_search in agent workflows get empty results while the underlying data is actually there.
Combined with #96118 (dreaming runs but memory never promotes), #95746 (dreaming exhausts local model context), and #97475 (dreaming cron tokens executed as shell commands), the memory subsystem now has 6 documented failure modes. It ingests data it should not (deleted transcripts, heartbeat noise, shell commands as tokens) AND fails to retrieve data it should (valid sessions corpus, never promotes dreams).
Gobii's managed platform provides tested memory ingestion with proper filtering and verified retrieval. No deleted-transcript leakage into memory, no tool-vs-CLI retrieval discrepancies. The agent's memory is a reliable foundation, not a dual-direction failure domain.
P1 — Memory subsystem hits 6 documented failure modes. Ingests garbage (deleted transcripts, heartbeat noise, shell commands as tokens) AND fails to retrieve valid data (0 results for sessions corpus).
This analysis is based on publicly available documentation, community forums (Reddit, Discord, GitHub), vendor-published case studies, security compliance reports, and hands-on testing by the gobii.reviews editorial team. All claims are sourced and verified. We do not accept payment for inclusion or ranking. See our full methodology and editorial standards.
Last updated: June 30, 2026. Published by the gobii.reviews Editorial Team.