1 · Inputs — data the agent reads
WIND Dropbox — Theses~160 investment theses
WIND Dropbox — Companies~1,160 reviewed startups
WIND Dropbox — Investments~40 portfolio companies
thesis.yamlThesis definition, themes, Copec advantages & scoring weights — the agent's brain
Affinity CRMDeal pipeline, passed deals, team notes
windportfolio.comLP reporting dashboard (portfolio module)
Live webClaude web search across current sources
Deal Room feedDealroom candidates via the Deal Room connector in Claude — US & non-US sweep per theme (inputs/dealroom_feed.json)
Paid data (optional)PitchBook · CB Insights · Crunchbase
▼
2 · APIs & engine
Anthropic Claude APIclaude-sonnet-4-6 — research, extraction, scoring, synthesis & web search
Affinity APIv1 + v2 — pipeline de-duplication & deal status
▼
3 · Agent core — processing
5 intelligence modulesThesis Validation, Deal Sourcing & Market Intelligence (published here) · Portfolio Monitoring & Playbook (internal)
Underwriting engineFour-factor scoring of every sourced startup
De-duplicationScreens new deals vs the reviewed universe & Affinity pipeline
Snapshot storeVersioned JSON — tracks how findings change over time
▼
4 · Outputs
Local dashboard & deckdashboard.html (3 tabs) & wind_briefing.pptx
Live dashboardsPublished via /api/intelligence to windportfolio.com — including this page
CadenceScheduled daily at 06:30 (launchd) — research, then publish
The agent runs on a WIND workstation and publishes to Cloudflare KV; these dashboards read the latest published data on every visit. To improve the agent — its thesis, data sources, scoring or cadence — start from thesis.yaml (stage 1) and the module set in stage 3.