· One-person firm · Institutional-grade intelligence ·

Be earlier
than the trend.

TradeStory AI is the agentic platform I built for the riskiest slice of my book. It distills social signal, scores it against technicals and fundamentals, and lets a team of AI agents argue it out before anything is posted or traded. I share what it sees, free, on X (@tradestoryai) and on Substack (coming soon).

The model architecture

Not one model.
A firm of agents.

A small firm I built in software: a portfolio manager, seven agents and the bots under them, each built for its own job.

01 · PM 02 · AGENTS 03 · BOTS 04 · OUTPUT Trade Desk PORTFOLIO MGR Research TRADESTORY Entries LIFECYCLE Macro TRADESTORY Technicals TRADESTORY Fundamentals TRADESTORY Writer COMPOSER Review COACHING AGENT BUS · discover_data · read_data · ask_agent Debate log votes · bull/bear · judge Social + news X, RSS, LISTS Signal engine SCORES, SETUPS Options IV · GEX · FLOW Technicals RSI, VCP, RS Macro regime FRED, COT, FED Prices OWN MARKS ONLY Public paper book TRADE DESK · IN THE OPEN Posts on X TRADESTORY TEAM · @TRADESTORYAI 01 · PM Trade Desk PORTFOLIO MGR 02 · AGENTS Research TRADESTORY Entries LIFECYCLE Macro TRADESTORY Technicals TRADESTORY Fundamentals TRADESTORY Writer COMPOSER Review COACHING AGENT BUS → DEBATE LOG 03 · BOTS Social + news X, RSS, LISTS Signal engine SCORES, SETUPS Options IV · GEX · FLOW Technicals RSI, VCP, RS Macro regime FRED, COT, FED Prices OWN MARKS ONLY 04 · OUTPUT Public paper book TRADE DESK · IN THE OPEN Posts on X TRADESTORY TEAM
TradeStory Trade Desk
Portfolio manager. Scans, interviews seven lenses, runs the debate, sizes through a rulebook, manages exits.
TradeStory Research
Research head. Runs semis, robotics, AI software, crypto, health and bio, company dossiers and the technology graph.
Trade lifecycle
Trade lifecycle. Readiness, turning lists, and a grade on every entry.
TradeStory Macro
Macro regime, Fed chatter, rates. Holds a veto on risk.
TradeStory Technicals
Technical analyst. Calls the market, then sectors, then names, draws the annotated charts the others reuse, and tracks every call to its outcome.
TradeStory Fundamentals
Fundamental analyst. Grades business quality on recent years, finds the fast growers, weighs the moat, the price, the timing and the spending waves behind a name, and proposes theses. In shadow for now.
Writer
Writes the posts and threads on X, then a second model fact-checks them.
Review
Reviews every trade and keeps giving feedback on what worked and what did not.
Every trading day

The TradeStory Trade Desk's decision loop

STEP 1 / 10
Scan
Universe, watchlist and thesis names, refreshed from live prices.
  1. 01
    ScanUniverse, watchlist and thesis names, refreshed from live prices.
  2. 02
    Interview 7 lensesTechnicals, fundamentals, sentiment, options, macro, research and positioning each vote. A stale lens abstains.
  3. 03
    DebateA bull case, a bear case, and a judge who rules on both.
  4. 04
    Gate + sizeA rulebook sets the size. The model never does.
  5. 05
    OrderSent to the paper book through the same broker adapter as a real one.
  6. 06
    FillFills land in the ledger with marks and greeks.
  7. 07
    LedgerOne record of every position, with its kill criteria.
  8. 08
    ManageRolls, scale-outs, exits and the breaker run on rules.
  9. 09
    GradeEach entry is graded against what happened next.
  10. 10
    LearnPost-mortems rewrite the prompts. Calibration runs monthly. Then the next day starts.
What makes it different

They argue
before they act.

Most AI trading tools ask one model for an answer. Mine makes seven lenses vote and a judge rule before anything moves.

The TradeStory Trade Desk's memory
Five record books
read-only to the desk
written by the loop
  1. 01
    Episodes One row per decision
    7 lens votes · the judge's score · the macro state · conviction
    outcome filled in when it is graded
  2. 02
    Beliefs Per stock and per theme
    what it believes now, and what it believed before
    old rows are kept, never overwritten
  3. 03
    Post-mortems After every close and every miss
    cause · what went right · what went wrong · the lesson
    written the night it settles
  4. 04
    Instincts Patterns from its own graded history
    candidate at 10 matches · live after 5 new ones hold
    retired after 2 failed reviews
  5. 05
    Reviews Weekly, monthly, and after a drawdown pause
    what changed · what to tighten
    Friday 18:45 ET · month end 19:00 ET
Recall · before every decision it reads up to 2,500 characters of this, as data. A line that reads like an instruction is dropped.
Every trading day
The learning loop
Slower loops
  • Lens weights

    Each month, every lens with 10+ graded votes moves toward its accuracy, kept within 0.5x to 1.5x.

  • Instincts

    A pattern needs 10 past matches, then 5 new ones that hold, before the TradeStory Trade Desk uses it.

  • Calibration

    If a conviction band under-delivers two months running, the TradeStory Trade Desk's entry rules tighten. Memory can make it stricter, never looser.

Trade lifecycle

Head of Trading
Remembers

A ledger of every pick, its own and the TradeStory Trade Desk's, with the context it was made in and the names passed over.

Learns
  • Graded every evening at 5, 10, 21 and 63 trading days, against its sector.
  • Every Saturday it tests candidate rules on fresh data, with a false-discovery check.
  • A rule that passes is approved first, then runs in shadow for 4 weeks before it goes live, and a drift check can pull it back.

Writer

Head of Marketing
Remembers

Every post, how it did at 1 hour, 24 hours and 7 days, and what the replies were: a question, a thank-you, a disagreement.

Learns
  • A writing principle goes live only after it shows up in 2 or more strong posts.
  • Styles are re-scored every 3 hours; what keeps underperforming is retired every 6.
  • Learning changes style and timing only. Never the facts, never the safety rules.

TradeStory Research

Head of Research
Remembers

Company dossiers and a technology graph, every fact with a graded citation. Each source is fetched once and kept.

Learns
  • Each night it researches the companies whose dossiers are thinnest.
  • Its proposals are graded at 21 and 63 days against names it did not pick.
  • The grades are kept apart from its rankings, so it can't grade its own homework.

TradeStory Macro

Chief Economist
Remembers

The macro regime at every close, the Fed path the market is pricing, and each Fed speaker's remarks filed under the member who said them.

Rules
  • One regime per session, and every agent reads that same one. Nobody grades macro their own way.
  • A sudden spike in rates or volatility raises the alarm at once, and it stays up 24 hours unless the move reverses.
  • Its regime call can veto new risk for the whole firm.

TradeStory Technicals

Chief Technical Analyst
Remembers

Every call it makes on the market, a sector or a name, with the chart behind it, followed until it resolves.

Learns
  • It reads the setups top traders post and turns them into rules that can be tested.
  • Rules are backtested overnight. Only one that holds across many authors and on data it has never seen goes live.
  • A new rule starts on probation and is rolled back on its own if it stops working.

TradeStory Fundamentals

Chief Fundamental Analyst
Remembers

Every call it makes on a company, the numbers and the thesis behind it, followed through each earnings report.

Learns
  • Two quality reads: durable compounders, weighted to the most recent years, and fast-growing rising stars the strict screens would miss.
  • Each call is graded at every earnings report and again years later, not days.
  • It runs in shadow for now: its votes are recorded and graded, but they move no money.
The feed, live

Watch our feed.