Fintech EM built an open-source AI wealth agent with Claude Code: 300 tools, 5 months
baggaarm · reddit · 2026-11-10
A Singapore-based fintech engineering manager released Open Worth (AGPL, free): a self-hosted AI wealth agent named Lucius, built with Claude Code over 5 months on a Claude Ultra subscription he routinely maxes out weekly.
What it is: essentially an agent harness — 300 tools, skills, automations, memory, arbitrary MCP connectors and artifacts, with per-tool permission levels; broker tools ask first by default. It logs spending daily from bank emails (works with scanned PDFs, no Plaid needed), sends Telegram morning briefings, checks prices every minute and judges news against your written investment thesis, places trades via IBKR/Coinbase/Gemini MCP servers (monthly VOO DCA automation), and all numbers come from code, never the model.
Building with Claude Code: Claude drafts specs, technical design and UI before any code; sticky decisions become ADRs; subagents split implementation, code review and security review; 20,000 unit/integration tests plus LLM-driven UAT where Claude writes acceptance tests and clicks through the real UI with screenshot proof; a hook refuses commits containing personal financial data.
Harness lessons: store exact approved arguments and re-run the saved call rather than trusting the model; the model won't finish automation work on its own — scheduling, retries, approvals and completion checks belong in the harness; the model is bad at math, so every balance comes from a ledger-backed tool; read open-source agent codebases like Codex to learn harness design; build a UI design system early.
More from coding & agent
- CopilotKit Open-Sources OpenBot 1.0, a Self-Hosted Personal AI Coworker — techNmak · 2026-11-10
- CopilotKit open-sources OpenBot 1.0, a self-hostable AI coworker platform with 11K stars — techNmak · 2026-11-10
- safeagents live RL runs: decision models cut false blocks toward reward-hacking-free training — yb2698 · 2026-11-10
- safeagents judge bench v2 leaderboard: Cloudflare clef-flash tops at 0.893 — yb2698 · 2026-11-10
- safeagents judge bench v2: 661 labeled agent tool calls to score safety judge models — yb2698 · 2026-11-10
- safeagents judge bench lands on HF: official leaderboard for decision-model safety judges — yb2698 · 2026-11-10