deepseekharness
An open-source AI memory management plugin that adopts a three-tier memory structure (stable zone, to-do zone, archive zone), supports weight-based eviction and persona locking mechanisms, and emphasizes data reliability and reproducibility.
Timeline (last 90 days)
deepseekharness is an open-source memory plugin whose core selling point is dividing memory into three layers: the stable zone, the to-do zone, and the archive zone.
deepseekharness's stable zone stores long-term settings, the to-do zone manages temporary tasks with deadlines, and the archive zone saves conversation summaries.
When the stable zone of deepseekharness reaches 100 entries, the oldest entry is evicted by weight rather than hard-deleted.
deepseekharness's core persona entries are never evicted under any weight calculation.
External testing conducted three rounds totaling 39 tests on deepseekharness, with an issue list accumulating more than 20 items.
When deepseekharness fails to read a memory file, it does not silently return an empty result, but instead explicitly reports an error or returns an error marker.
deepseekharness supports concurrent writes from four processes without data loss.
deepseekharness keeps three snapshots and supports rollback.
According to the data provided by the uploader, injecting 1000 snapshots into deepseekharness takes 76 milliseconds and uses 0.5 megabytes of disk space.
deepseekharness's performance figures can be reproduced via scripts, and it has zero dependencies, running with a single command after cloning.
All Facts (8)
deepseekharness is an open-source memory plugin whose core selling point is dividing memory into three layers: the stable zone, the to-do zone, and the archive zone.
50%Unverifieddeepseekharness's stable zone stores long-term settings, the to-do zone manages temporary tasks with deadlines, and the archive zone saves conversation summaries.
50%UnverifiedWhen the stable zone of deepseekharness reaches 100 entries, the oldest entry is evicted by weight rather than hard-deleted.
50%Unverifieddeepseekharness's core persona entries are never evicted under any weight calculation.
50%UnverifiedWhen deepseekharness fails to read a memory file, it does not silently return an empty result, but instead explicitly reports an error or returns an error marker.
50%Unverifieddeepseekharness supports concurrent writes from four processes without data loss.
50%Unverifieddeepseekharness keeps three snapshots and supports rollback.
50%Unverifieddeepseekharness's performance figures can be reproduced via scripts, and it has zero dependencies, running with a single command after cloning.
50%