Substrate

The AI value chain, as a connected graph — silicon → inference/agents.

An equity research tool that maps the entire AI supply chain — raw materials to chips to datacenters to the applications on top — as a graph of layers, not just a list of tickers. Built to dissect where momentum and stress flow between layers, and to spot buy/sell setups from price action, informed by your own strategy, your portfolio, and Claude's insights.

Built as a supporting research tool alongside Robinhood. The specialization is layer health: surfacing cross-layer dynamics, correlation breaks, and volume anomalies that aren't visible in a standard brokerage. A Robinhood MCP integration (agentic account access) is on the roadmap.

---

Architecture

                ┌──────────────────────────────┐
                │  topology.yaml (repo root)   │
                │  70 tickers · 9 layers ·     │
                │  93 supply chain edges       │
                └──────────────┬───────────────┘
                               │
              ┌────────────────┴────────────────┐
              ▼                                 ▼
   ┌─────────────────────┐         ┌────────────────────────┐
   │  Python agent       │         │  Next.js dashboard     │
   │  agent/refresh.py   │         │  (this app)            │
   │  agent/insights.py  │         │                        │
   └──────────┬──────────┘         └──────────┬─────────────┘
              │                                │
              │   writes                       │   reads
              ▼                                ▼
        ┌─────────────────────────────────────────────┐
        │            Supabase (Postgres)              │
        │  tickers · prices · holdings · insights ·   │
        │  refresh_log · conversations · messages     │
        └─────────────────────────────────────────────┘

Invariants:

• The Python agent is the only writer to tickers, prices, insights, refresh_log
• Next.js writes only to holdings, conversations, messages
• ANTHROPIC_API_KEY is server-side only — never exposed to the browser

---

Features

View	What it does
Pipeline graph	Left-to-right SVG of the full AI supply chain, colored by today's performance
Layer heatmap	All 70 tickers grouped by layer, heat-colored by 1d/5d/1m/YTD return
Pullback radar	High-beta names sorted by distance from 52w high, RSI, and MA signals
Correlations	Layer-to-layer 90d correlation matrix + recently decoupled pairs
Anomalies	Layer divergence z-scores, volume spikes, detected signals
Portfolio	Holdings tracker with layer allocation donut and P&L
Trade journal	Trade thesis log with options tracking (strike, expiry, ITM/OTM)
Daily brief	On-demand EOD report: layer health, movers, portfolio, open setups
Signal feed	Chronological stream of detected anomaly / flow / signal insights
Ask Claude	Conversational analyst grounded in topology + live prices + your positions

Plus config views (Topology, API keys) and a per-ticker detail page (/ticker/[symbol]) with a YTD chart, key stats, and its upstream/downstream neighbors.

---

Stack

• Frontend: Next.js 16 (App Router) + TypeScript + Tailwind CSS
• Database: Supabase (Postgres)
• Data pipeline: Python ≥3.13 + uv — agent/refresh.py pulls EOD prices via yfinance
• AI: Anthropic SDK — claude-sonnet-4-6 for reports and Ask Claude, claude-haiku-4-5-20251001 for routine anomaly/flow detection narratives
• Charts: Recharts. Pipeline graph is hand-built SVG.
• Icons: Lucide React

No shadcn/ui, no react-flow, no component libraries — Tailwind primitives only.

---

Setup

1. Create Supabase project

Create a free project at supabase.com. Copy the project URL, anon key, and service role key.

Run the migrations in the Supabase SQL Editor (in order):

-- supabase/migrations/*.sql

2. Configure environment

cp .env.local.example .env.local
# Fill in NEXT_PUBLIC_SUPABASE_URL, NEXT_PUBLIC_SUPABASE_ANON_KEY,
# SUPABASE_SERVICE_ROLE_KEY, ANTHROPIC_API_KEY

3. Install and run

npm install
npm run dev

predev/prebuild automatically regenerate the topology and watchlist TypeScript from topology.yaml — no manual codegen step needed.

4. Initial data load

cd agent
cp .env.example .env
# Fill in SUPABASE_URL, SUPABASE_SERVICE_ROLE_KEY, ANTHROPIC_API_KEY

uv sync
uv run python refresh.py -v --period 2y   # initial backfill (~2 min)
uv run python insights.py                  # generate first insights

5. Schedule daily refresh

# Weekdays after market close (adjust timezone)
0 17 * * 1-5 cd /path/to/substrate/agent && uv run python refresh.py --period 1mo
30 17 * * 1-5 cd /path/to/substrate/agent && uv run python insights.py

---

Topology

The supply chain is defined in topology.yaml — 9 layers, 70 tickers, 93 directional edges encoding real supplier relationships. Edit this file to add tickers or adjust the graph. The TypeScript types are auto-generated at build time via scripts/gen-topology-ts.mjs.

Layers: Raw Materials → Energy → Storage → Foundries/Memory → Chip Design → Quantum → Photonics → Datacenter → Applications

---

Cost

• Supabase: Free tier (500MB, more than enough)
• yfinance: Free (EOD prices)
• Anthropic: ~$2–5/month with daily briefs + moderate Ask Claude usage

---

Disclaimer: Not financial advice. For personal research use only.