Roadmap
Memory Layer's engine is strong: hybrid retrieval, an ACT-R-grounded activation model, evidence-backed validation, memory consolidation into insights, a code graph, an automation-loop control plane, and a real evaluation harness.
This roadmap is about the other half — making it easy for a much wider audience to understand, use, and install: from a first-time beginner to a research scientist, from a classroom to a 30-second demo clip to an unconventional integration.
Two principles guide it. Prefer approaches with scientific backing or a tried-and-tested precedent, and keep a few deliberately experimental bets to stay fresh. And sequence foundation-first: an effortless install and a first-run "wow" unlock every other audience.
The nine tracks
A clean, legible backlog and this public roadmap. Land first.
One command to a working system: a full Docker Compose stack, a one-line installer, a smarter memory doctor, a single-step setup, and real Windows support. The funnel gate for everyone.
memory demo seeds a showcase project so query, graph, and resume show something immediately — plus a guided first-run tour and a five-minute quickstart. Grounded in the worked-example effect: people learn a system faster from a filled one.
A "three things to know" mental model, a glossary, and docs restructured into Beginner / Daily use / Advanced / The science.
The 3D memory graph as the hero visual — with activation rendered as colour and size, and an experimental animation of spreading activation and decay that literally shows the ACT-R model in motion.
A documented HTTP API with an OpenAPI spec, Python and TypeScript clients, non-git ingestion, and recipes — Obsidian, a research corpus, game NPC memory, home automation, and memory for your AI chats.
Turnkey one-command reproduction, citeable experiment bundles, an anonymized dataset with a DOI, and a methods write-up.
A keyless single-machine classroom pack, a student mode, and a cognitive-science curriculum taught with the live visualization as the instrument — the tool teaches the concepts it implements.
Green the adversarial answer-quality gate, fix known bugs, make every error name its own fix, and add opt-in usage telemetry so we can tell whether the funnel improved.
What we start with
The first five: reconcile the backlog, ship a user-facing Docker Compose stack, build memory demo, fix a docs/code divergence in the memory-type list, and green the adversarial answer-synthesis gate — plus publishing this roadmap.
Deliberately experimental
- A zero-dependency embedded mode for demos and classrooms (a spike first — we will not compromise the Postgres+pgvector engine to get there).
- Animated activation, grounded in the model the engine already runs.
- The tool as a teaching instrument — turning the cited science into classroom lessons driven by the live graph.
Not yet
Hosted/SaaS, internationalization, and a hosted multi-tenant sandbox are deliberately deferred until the API is stable and demand is measured.
The full, tracked roadmap — with per-ticket scope, reuse pointers, and estimates — lives in docs/roadmap.md in the repository and in the Memory Layer Linear project.