cupel · v0.1 · MIT
A personal investing analyst
cupel turns the LLM in your AI harness into a research companion that learns your edges and tests your ideas against the investing canon. It lives in one folder and keeps a dated record of your thinking. It gives reasoned, mandate-grounded calls — the likely scenarios, rough upside and timeframe, and the risks named. It refuses false precision dressed as fact and bare tips with no reasoning, and never places trades; the decisions stay yours.
Use what you already know
The premise is Peter Lynch’s: your edge is what you already know from your work and daily life — the products you watch take off, the industry shifts you see before the analysts do. A familiar name is only a lead, though; it still has to survive the homework.
cupel turns the LLM in your AI harness into a personal research analyst that captures those edges, researches an idea properly, and keeps your reasoning honest as your thinking accumulates in one place.
cupel lives in one folder, your office, and accumulates everything it learns about you there.cupel sharpens your own reasoning, surfaces what you are missing, and recommends what it would do and why — the likely scenarios, rough upside and timeframe, with the risks named. Investing is a bet; cupel helps you make an informed one. It refuses false precision dressed as fact and bare tips with no reasoning, and never places trades; the decisions stay yours.
The stance
Edge before opinion. A familiar product or a hot tip is only a lead; it still has to survive the homework. A wonderful business at the wrong price is a poor investment. Most of a normal person’s money can sensibly sit in a low-cost index; individual picks are a considered satellite, only where you have a real edge.
Set up the office, add the skill
Install the package:
$ npm install -g @samgalanakis/cupel
Set up your office, the vault where cupel keeps everything it learns. It defaults to ~/cupel; override with CUPEL_HOME:
$ cupel init
Add the skill to your harness
cupel ships as a skill for Claude Code, Cursor, Gemini CLI, Codex, OpenCode, Kiro, Pi, Qoder, Trae, and GitHub Copilot. Install auto-detects which harness directories exist:
$ cupel skills install # auto-detect installed harnesses
$ cupel skills install --all # every supported provider
$ cupel skills update # re-sync after npm update -g @samgalanakis/cupel
$ cupel skills check # installed version + content-hash status
Then talk to it from any harness: /cupel.
Your first session. Say /cupel and, on a fresh office, it interviews you, writing your profile, edges, mandate, and a few trusted sources. Then hand it an edge-driven idea, like /cupel I keep seeing this product everywhere at work: it resumes any prior notes, researches the company, runs the discipline gate, and files a thesis with explicit falsifiers. Come back with /cupel brief any time for a status check. cupel reads and writes your office; you make every call.
Without the npm package, the skill is committed at skills/cupel/, so you can pull it straight from GitHub:
$ npx skills add SamGalanakis/cupel
Or add it as a Claude Code plugin marketplace:
/plugin marketplace add SamGalanakis/cupel
/plugin install cupel@cupel
Claude Cowork
Cowork installs plugins from a zip, not from npm. Every release ships a cupel-cowork.zip asset that bundles the CLI with zero npm dependencies, so cupel init and cupel doctor run inside the sandbox on node alone.
npm run build:cowork.
One folder that remembers everything
cupel keeps all of its state in one Obsidian-compatible vault, the office, never in the current project. The vault is plain markdown with YAML frontmatter for structure and [[wikilinks]] for provenance: a thesis links its [[source]]; a position links its [[thesis]].
Open it in Obsidian for the graph view, or just read the markdown. Keep it under git: the decision journal’s value compounds as its history grows. Read the full office schema, including note types, the cupel doctor linter, and the journal discipline.
Talk to it, or go direct
Say /cupel and it figures out what you need: it locates your office, takes its pulse, and routes you to the right job. Or name the command yourself.
cupel doctor.pulse plus the executive readout: what changed, what needs attention, next moves.When you name a specific company, cupel runs cupel show <TICKER> first to resume exactly where you left off. It never starts cold on something the office already knows.
The filing cabinet, kept honest
The cupel binary does the deterministic bookkeeping so the LLM can do the judgment. It manages the office, validates it, and installs the skill.
~/cupel (or CUPEL_HOME).cost-basis to last-price and a size-weighted price return — and, once capital is set, value and gain in money. Recorded prices, never live quotes.portfolio turn allocation percentages into money. cupel never fetches account balances; you maintain it.cupel stamp pulse) so the front door knows how fresh the office is.Division of labour
The LLM does the judgment; the linter guards the filing cabinet. cupel doctor catches a satellite over your max-position-pct (the core is exempt), the satellite total over its target, a large idle-cash remainder, a holding unreviewed for months, a decision whose review date has arrived, or a source long unchecked, and the companion surfaces it before you act.
Anchored in the classics
cupel’s judgment is anchored in a small canon chosen for accessible, individual investing: Lynch and Mayer on edge, Dorsey on moats, Housel and Bernstein on temperament and risk, Bogle on humility, Graham and Marks on price. The audience is a smart person who is an expert in some other field, so the principles stay plain and free of jargon.
Read the full canon: eight thinkers as operating lenses, plus the six Lynch categories that drive the story, the numbers, and the sell signal.