AILANTA
All research
GitHub · npm · PyPI · Discovery evidence

AI memory is becoming an auditable data layer

A GitHub and package-registry cohort tests whether persistent agent memory is acquiring provenance, retention, deletion, access-control, and poisoning defenses.

2026-08-12266 items10 source types20 validation posts
Main finding

Early evidence; the market is not yet confirmed

266 subject-filtered implementations show that persistent memory is already a visible software category. Persistence appears in 15.4% of the cohort, but provenance or audit in 3.8%, retention controls in 1.5%, deletion in 2.3%, and poisoning defenses in 0.4%. Memory supply is growing faster than memory governance.

01

Market snapshot

Comparable measurements from independent market surfaces.

266

Implementations and packages

Deduplicated and subject-filtered primary cohort.

111

Median repository stars

Calculated across repositories with at least one star.

3.8%

With provenance or audit

10 items in the subject-filtered cohort.

0.4%

With poisoning defenses

1 items in the subject-filtered cohort.

02

Search-match dynamics

Bars show monthly GitHub query matches across persistent memory and governance terms; loaded counts are deduplicated implementations retained for feature analysis.

Jan 2641318 loaded
Feb 2693828 loaded
Mar 26125829 loaded
Apr 26109631 loaded
May 2694435 loaded
Jun 2684241 loaded
Jul 2661132 loaded
Aug 2610822 loaded
03

What exists inside the category

One item may contain more than one feature.

Persistent memory stores

41 · 15.4%

Provenance and attribution

10 · 3.8%

Retention and expiration

4 · 1.5%

Deletion and forgetting

6 · 2.3%

Memory access control

4 · 1.5%

Memory poisoning defenses

1 · 0.4%

Influence tracing

7 · 2.6%

04

Representative projects

05

Cross-source validation

These publications are not part of the primary numeric cohort.