Classroom

Memory Layer in the classroom

Memory Layer implements ideas lifted directly from the cognitive science of memory — ACT-R activation, exponential forgetting, spreading activation, complementary learning systems. That makes it an unusual teaching asset: the software is a working model of the theory, and the live graph visualization is the lab instrument. Students don't read about forgetting curves; they watch nodes cool down.

Three lessons, each ~45 minutes, each built around observable behavior:

The classroom pack

Everything runs on one machine with Docker, no cloud, no API keys, and no per-student accounts. The pack lives in the repository under classroom/: a compose overlay, a seed script for the shared exercise project, a teacher README, and a printable student worksheet.

docker compose -f compose.yaml -f classroom/compose.classroom.yaml up -d
./classroom/seed.sh
# students open http://localhost:4040, project "classroom"

Two class modes:

  • Shared read-only set (lessons 1–2): start the stack with CLASSROOM_READ_ONLY=true and the service enters student mode — every write is rejected with a clear message, the web UI shows a banner, and nobody can break the exercise set. Queries still reinforce activation, deliberately: the class collectively heats up the graph, which is the observable at the heart of lesson 1.
  • Per-student sandboxes (lesson 3): writes on, one project per student (./classroom/seed.sh alice). Projects are isolated namespaces.

Why keyless matters here

In keyless mode retrieval is lexical and answers are extractive with citations and honest refusals. Nothing needs an account, no student data leaves the machine, and the behavior is deterministic — the same query produces the same evidence, which is what you want in a lesson. LLM synthesis is an optional upgrade, not a prerequisite.

Prep checklist

  • Docker installed, repository cloned, pack seeded (10 minutes).
  • Projector showing the web UI Graph tab, project classroom.
  • worksheet.md printed, one per student or pair.
  • Skim the reinforcement and consolidation pages — every mechanism the lessons use is documented there with its citations.

© 2026 Olivier Van Acker (3vilM33pl3). Memory Layer is AGPL-3.0-or-later with commercial licensing available.

On this page