My favorite project. Not because it’s the most technically interesting — because it solves the most honest problem. Estate and Trust management require my reasoning. When I’m unavailable, my family lacks access to my thinking on the decisions that matter most. Generic AI can’t substitute for a specific person’s values and frameworks.
The Problem
The family’s most consequential decisions depend on judgment I’ve built over years. A Trust, an investment call, a major life decision — these need a perspective that understands the context, not just the question. When I’m not available, there’s no way to access that perspective without me.
What It Is
A RAG system that ingests my writing — ClemVault documents, recorded decisions, documented reasoning — into a local system prompt. My wife asks questions in natural language. The advisor responds in my voice and cites the source documents behind each answer. All context stays local. Nothing goes to a cloud database.
This is not Clembot. Clembot is a client tool. This is me — or the closest approximation of my reasoning available when I can’t be reached.
Features
- Clemens persona loader ingested from ClemVault documents
- Trust-specific advisor mode with estate context
- Local-only storage (no external data beyond the LLM call)
- Calibration interface for adding reasoning entries over time
- Source attribution on every answer
Stack
| Tool | Role |
|---|---|
| Claude API (or Ollama) | Language model — Claude for quality, Ollama for full privacy |
| Hono | Local API server |
| React + TypeScript + Tailwind | Chat interface |
| Supabase (optional) | Encrypted remote sync |
| Cloudflare Pages | Local network hosting |