memgovern
An open-source AI agent memory governance tool focused on the governance layer for memory writes and deletions. It provides mechanisms including decay-based forgetting, tombstone marking, conflict arbitration, and a trust ledger. The core package has zero dependencies and is built on SQLite.
Timeline (last 90 days)
memgovern 0.5.0 added the `memgovern audit` command-line tool, which outputs a human-readable trust report in the terminal.
memgovern is a governance layer that focuses on the governance of memory writes and deletions, rather than memory storage.
The memgovern core package has zero dependencies and uses only the standard library, with SQLite as the underlying backend.
memgovern uses tombstone markers for soft deletion rather than physical deletion to ensure traceability.
memgovern has maintained a per-source trust ledger since v0.2.
memgovern's trust scoring uses a Bayesian-smoothed win rate with an initial prior of 0.5 and a 30-day half-life decay.
A memgovern trust score below the threshold of 0.35 triggers a LOW annotation.
memgovern sets three types of tripwires: injection marker detection, burst write detection, and low-trust sources challenging high-trust holders.
memgovern adopts a quarantine-before-arbitrate design philosophy, where writes that hit a tripwire are quarantined pending manual review.
Before v0.5, the trust ledger could only be accessed via the Python API (store.trust_report()) or the memory_trust_report MCP tool.
4 more timeline events
All Facts (14)
The author acknowledges that memgovern is a structural defense rather than a semantic defense, that attackers can gradually raise their trust score through sustained benign writes, and that the marker list will miss rewritten variants.
50%Unverifiedmemgovern is a governance layer that focuses on the governance of memory writes and deletions, rather than memory storage.
50%UnverifiedThe memgovern core package has zero dependencies and uses only the standard library, with SQLite as the underlying backend.
50%Unverifiedmemgovern uses tombstone markers for soft deletion rather than physical deletion to ensure traceability.
50%Unverifiedmemgovern has maintained a per-source trust ledger since v0.2.
50%Unverifiedmemgovern's trust scoring uses a Bayesian-smoothed win rate with an initial prior of 0.5 and a 30-day half-life decay.
50%UnverifiedA memgovern trust score below the threshold of 0.35 triggers a LOW annotation.
50%Unverifiedmemgovern sets three types of tripwires: injection marker detection, burst write detection, and low-trust sources challenging high-trust holders.
50%Unverifiedmemgovern adopts a quarantine-before-arbitrate design philosophy, where writes that hit a tripwire are quarantined pending manual review.
50%UnverifiedBefore v0.5, the trust ledger could only be accessed via the Python API (store.trust_report()) or the memory_trust_report MCP tool.
50%UnverifiedThe memgovern audit command supports --json for machine-readable output and --source NAME for filtering a single source.
50%Unverifiedmemgovern can be installed via pip install memgovern.
50%Unverifiedmemgovern is open-sourced under the MIT license and is available on GitHub (hahahahahahahahah6/memgovern) and PyPI.
50%Unverifiedmemgovern 0.5.0 added the `memgovern audit` command-line tool, which outputs a human-readable trust report in the terminal.
50%