Agent Human Memory · Planned OSS release

Explore agent continuity without turning memory into a black box.

Agent Human Memory is an experimental project for explicit local records such as preferences, goals, episodes, and working context while keeping provenance, authority, review, retention, and forgetting visible.

Release roadmap Compare the stack
Explicit recordsMemory should remain inspectable instead of becoming hidden model state.
ForgettableReview, retention, and forgetting are part of the lifecycle design.
Safety before recallAuthority, provenance, consent, and false merges matter more than maximizing retrieval.

What the project is designed to explore

Continuity

Which explicit memory kinds can reduce repeated instructions across sessions without creating hidden state?

Authority

Can the system preserve the difference between context, preferences, and delegated instructions?

Forgetting

Are review, retention, and forget controls understandable enough for users to remain in charge?

Consolidation safety

How can recurrence and consolidation avoid unsafe false merges or provenance escalation?

Safety boundary: this project prioritizes understandable authority, provenance, and forgetting over maximizing how much an agent remembers.

Public release status

Agent Human Memory is planned for public OSS release. The experimental implementation remains behind a private repository while public packaging, documentation, licensing, and safety-oriented release gates are finalized. Public repository and installation links will appear only after the release is ready.

Product boundary

This is an experimental memory infrastructure project, not an identity-simulation product. It is intended to make agent memory more inspectable and controllable, not to infer hidden traits.