Case study / 07
Governed AI
Memory Prototype
A governed AI memory system where knowledge ingestion from web, PDF, and paper sources gets reviewable structure instead of loose AI summaries.
Type
Knowledge system
Platform
Local prototype
Status
Governed AI memory prototype
Stack
Next.js · Postgres · Playwright · Docker
Overview
Knowledge ingestion from the web, PDFs, and papers needs reviewable structure, not loose AI summaries.
Mnemos uses a Next.js UI, a Node.js control plane, and a Playwright executor with Postgres as source of truth and an LLM patch-review protocol for auditable updates.
System architecture
UI
Next.js
Control
Node.js control plane
Ingest
Playwright executor
Truth
Postgres source of truth
Review
LLM patch protocol
Ops
Docker Compose + health checks
Problem
AI summaries drift. Knowledge captured from sources needs structure, provenance, and review before it becomes memory.
What it does
- Web / PDF / paper ingestion
- Patch protocol
- Audit logs
- Health checks
- Topic-based notes
What this proves
- Governed AI memory design and auditability.
- LLM patch-review architecture over direct writes.
- Playwright-driven ingestion and Dockerized local AI systems.
- Health-check thinking for local infrastructure.
Note
The repository is private. This public case study is sanitized; source and internal implementation details are not published.