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
E0 · Backlog & roadmap foundation
A clean, legible backlog and this public roadmap. Land first.
E1 · Effortless install
The Docker Compose stack, one-line installer, memory wizard, and memory doctor form the current setup path. The remaining work is making that path clearer and more reliable across platforms.
E2 · First-run wow
memory demo, memory tour, and the five-minute quickstart already provide a filled, keyless first run. Keep improving the demo and its empty states without adding setup burden.
E3 · Concept coherence & docs
The essentials-first docs and optional deep dives make the main workflow approachable while keeping the mental model, glossary, and science available when wanted.
E4 · Demo & video assets
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.
E5 · Integration surface & API
The HTTP API reference, non-Git ingestion, and integration recipes are in place. Client libraries and additional integrations remain the next expansion path.
E6 · Research & reproducibility
Turnkey one-command reproduction, citeable experiment bundles, an anonymized dataset with a DOI, and a methods write-up.
E7 · Education & classroom
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.
E8 · Trust & polish
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.
Current baseline
The public setup, demo, quickstart, API reference, ingestion path, recipes, and documentation structure are available today. The remaining tracks describe improvements beyond that baseline, not steps required before trying Memory Layer.
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.