Documented API shape
The README exposes storage, search, and management flows clearly enough to prove this is a working system, not just a concept.
Back to selected work
Local AI infrastructure / semantic memory
Memory-Pi
A local vector memory layer for an ecosystem of AI-assisted services running close to the edge.
Domain
Local AI infrastructure / semantic memory
Role / ownership
Architecture, memory model, service orchestration, local AI integration, and the UI that developers actually use to interact with it.
Stack
Node.js · Fastify · TypeScript · sqlite-vec · Ollama · OpenRouter · Vite UI
01 — Challenge
Challenge
A useful local AI ecosystem needs memory, retrieval, ingestion, and operational simplicity. Most off-the-shelf AI demos do not solve that in a local-first way.
The system had to be practical, multimodal, and lightweight enough to support a broader Pi ecosystem.
02 — Solution & architecture
Solution & architecture
A local semantic memory service: sqlite-vec for vector search, local embeddings, structured ingestion, and a UI to capture and audit what the system remembers.
Its role inside the wider Pi ecosystem makes it especially strong: it is infrastructure, not just an isolated experiment.
03 — Tradeoffs
Tradeoffs
Built for practical retrieval and ingestion, not sci-fi agent demos. Less viral on Twitter, more useful at 2 AM when something breaks.
Kept the stack lightweight so it runs on a Pi without melting. That meant saying no to heavier vector databases and cloud dependencies.
04 — Outcome & proof
Outcome & proof
Ecosystem role
It is the memory layer the rest of the Pi services rely on. Without it, the other tools have no context.
05 — What it proves
What it proves
Pietro builds infrastructure, not just front ends.
He thinks in systems, lifecycles, and real-world use — not just surface features.