The kata for personal
capital stewardship.
A markdown-first, MIT-licensed substrate that turns any code agent — Claude Code, Cowork, Codex, Gemini CLI, OpenCode — into a sophisticated decision-support engine for personal wealth management.
Not a trading bot. Not a robo-advisor. Not alpha. The discipline-imposing scaffolding that turns intuition + checklists into outperforming intuition alone — borrowed from medicine (Gawande), aviation (cockpit cross-checks), and software engineering (RFC + post-mortem).
What this is — and isn't.
Most retail "AI trading systems" lose money to brokers, slippage, bias, and overconfidence. The compounding edge for a part-time investor is not making catastrophic decisions, more than alpha generation. This substrate is built for that, and only that.
What this is
- A decision-support framework that imposes structure on weekly decisions
- A discipline-enforcement scaffolding that catches biases through structured debate
- A journal that compounds — sessions accumulate into a learning corpus
- A tax-aware position planner that surfaces jurisdiction-specific implications
- An audit trail of past decisions with retrospective outcome tracking
What this is NOT
- NOT an alpha-generation system
- NOT a trading bot
- NOT a robo-advisor
- NOT financial, tax, or legal advice
- NOT a guarantee of returns
- NOT secure if you skip the security model
The four plausible outcomes (no guarantees)
After 12+ weeks of consistent use, the substrate plausibly delivers:
- One catastrophic decision avoided in a 5-10 year horizon
- One tax mistake caught before year-end
- One position-size error surfaced before you took it
- One biased pattern in your thinking identified through retrospectives
The atomic unit: the Strategy Session.
A closed, dated, JSON-schema-validated artifact that captures everything needed to make and remember a wealth decision. Every command, every loop, every agent ultimately produces or consumes one.
Sessions are append-only after settlement. Outcome backfilling happens 30/60/90/180 days later as a separate write event — preserving the audit trail.
The 11-agent strategy team.
Three layers (Analysis → Risk → Synthesis) plus a cross-cutting Researcher. Each agent is a structured prompt + tool budget + output schema, not a separately-trained model. Swap the underlying LLM per agent based on cost vs capability.
Analysis layer
5 agents- Macro-Risk
- Crypto-DCA
- DeFi-Yield
- Fundamentals
- Technical
Risk layer
3 agents- Risk-Manager
- Tax-Optimizer
- Regulatory-Risk
Synthesis layer
2 agents- Portfolio-Manager
- Chief-of-Staff
Cross-cutting layer
1 agent- Researcher (cited primary sources)
Three tiers. Start at Tier 0.
Tier 0 is bootable in 10 minutes with any agent CLI — no Python, no Docker, no API keys. Tier 2 is engineering effort. Most operators live at Tier 1.
Markdown only
- Audience:
- Anyone
- Tooling:
- Markdown + any agent CLI
- Time:
- 30 min/week
- Outcome:
- Strategy Session journal + DCA discipline + tax calendar
OSS install
- Audience:
- Intermediate
- Tooling:
- Multi-agent debate runtime, OpenBB feeds, Notion hub
- Time:
- 2 hr/week
- Outcome:
- Weekly debate transcripts + portfolio drift alerts
Live execution
- Audience:
- Advanced
- Tooling:
- AgentKit / Alpaca live, Freqtrade, custom MCPs, ReasoningBank
- Time:
- Engineering effort, ongoing
- Outcome:
- Autonomous DCA + thesis-tracked execution
Five operating loops.
Each loop runs at a fixed cadence with a defined output. Most operators run only Loop 2 (Weekly Review) at first; Loops 3-5 add value as the substrate matures.
Loop 1: Daily Pulse
Weekday morning · ~2 minOne-screen digest: top movers, renewals, due actions
Loop 2: Weekly Review
Sunday · 15-45 minThe heartbeat. Multi-agent Strategy Session. Run this if nothing else.
Loop 3: Monthly Close
1st of month · ~30 minNet worth snapshot, runway, subscription audit, Box 3 running total
Loop 4: Quarterly Tax & Strategy
Jan/Apr/Jul/Oct · ~45 minBox 3 vs BV analysis, BTW prep, cost basis reconciliation, security audit
Loop 5: Annual Architecture
December · ~2 hoursYear of sessions distilled. Pre-Jan-1 position optimization. Inheritance protocol freshness.
Skipping Loop 5 is the most common operator failure. The annual review compounds the value of every weekly session. Without it, the corpus accumulates without distillation.
Quickstart — 10 minutes.
Tier 0 setup is markdown + agent CLI. No infrastructure.
Clone
git clone the repo. Read README, architecture/00-overview.md, architecture/10-honest-limits.md.
Read one example
examples/jane-freelance/sessions/2026-W18.md (Tier 0) or examples/marcus-founder/sessions/2026-W18.md (Tier 1).
Initialize private store
Outside the substrate repo: ~/iis-private/{sessions,snapshots,theses,retrospectives,trajectories}.
Run first session
In your agent CLI, invoke /weekly-strategy. Provide portfolio context. Read draft. Mark verdict.
Sunday cadence (after first session): Run /weekly-strategy. Read the draft. Mark verdict. Schedule retrospective for any thesis-driven action (90 days). 15 minutes per week.
Why this exists.
A 28-year-old AI Architect in Amsterdam, building a living legacy for godchildren and nieces, found that the wealth-decision scaffolding he'd battle-tested privately was generalizable. The architecture (not the positions) is now extracted, sanitized, and published under MIT so other creators with similar profiles can boot the same discipline.
Sibling systems
IIS is part of a unified memory ecosystem. SIS is the substrate's substrate; Library OS is the book-intelligence sibling; IIS is the wealth-decision domain.
Discipline beats intuition.
Process beats prediction.
Audit beats optimism.
Internalize that, and the substrate gives you what it can give you. Reject it, and look elsewhere — there are better tools for whatever else you want.
MIT License. No financial advice. No alpha guarantee. Decision support only. Read honest limits before use.