- Category
- AI Agents
- Rank
- No. 3565Tools index
Previous survey · No. 3432 ·
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Agent Building
- GitHub
- 12 stars
- Date
About
SelMem is an experimental Rust memory system for LLM agents, backed by SQLite, that selectively encodes, distorts, and consolidates experiences over time instead of keeping a complete unmodified log. The project tests whether this kind of imperfect, path-dependent memory causes separate agent instances to diverge in behavior rather than converging on the same average response.
At a glance
SelMem is an experiment in how selecting and reconstructing past experiences affects a model’s later choices. It is worth reading if you are designing a memory evaluation and need examples of controls that distinguish recall from delivery.
Good fit
Researchers comparing agent-memory designs on synthetic histories.
Consider something else when
Choose a production memory store if your immediate problem is retrieving records reliably.
Start here
Read the matched full-memory and empty-memory conditions, then inspect which beliefs and episodes reached the model. Check identical-request disagreement before interpreting a difference.
Cost and requirements
MIT-licensed Rust repository with documented experiments. Reproducing a hosted-model run requires provider access.
Limits to know
The reported scenarios are synthetic and share a behavioral theme. Changes after recall do not establish a general creativity or real-world performance benefit.
Sources checked 2026-11-05. Source-based assessment.
Why it made the leaderboard
Most agent memory systems maximize recall; this tests the opposite bet, that selective forgetting could let agent instances develop distinct behavioral identities over time.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.