Stop screening five hundred names in a spreadsheet. Ask Warren — get a ranked shortlist and a written thesis on every top pick.
WARREN is an autonomous equity-research agent. It scans the S&P 500 and Nasdaq-100 (≈580 tickers), drops the obviously expensive and illiquid names with a rules-based pre-filter, scores every survivor 1–10 on undervaluation with a written reason, and writes a full thesis on the top ten — bull case, bear case, what would change the call, catalysts, risks. The whole process streams live to an operator console so you can watch the agent reason rather than wait for a list. Built on the AGENT X pipeline pattern; same operating shape as Claude Code, pointed at equities instead of code. Live soon at WarrenTrades.com.
Equity research is bottlenecked by attention. A good analyst can cover twenty names well, fifty competently, two hundred superficially. WARREN reproduces the work itself — universe definition, fundamentals pull, quantitative gate, qualitative score, full thesis on the top picks — and executes it across the entire S&P 500 plus Nasdaq-100 in a single scan. The investor reads the artifact and decides.
A single WARREN scan covers ~580 unique tickers across the S&P 500 and Nasdaq-100. No selection bias from a watchlist the analyst built two years ago. Names that wouldn't normally cross the desk get the same fundamental scoring as the household ones.
Not a star rating, not a numeric score in isolation — a full analyst-grade write-up on each top-ten name. Bull case. Bear case. What would change the call. Catalysts to watch. Risks that bind. The artifact you'd expect from a junior analyst working a name for a full afternoon.
The agent's work streams to the console step by step — which tickers passed the gate, why a name scored 8 instead of 6, the analyst's reasoning as it writes the thesis. The investor watches the agent think and can intervene at any step. Closer to pairing with an analyst than reading a finished report.
Scans are sold in packages — single scans, monthly bundles, or unlimited tiers — priced from roughly $25 to $50 per scan depending on tier. The investor can refresh a thesis on a name or rerun the full universe whenever market context shifts, without re-engaging an analyst or rebuilding a research budget. One package, the whole S&P and Nasdaq-100 on tap.
Every scan is cached with its inputs (universe, snapshot, prefilter thresholds, model versions) and its outputs. The investor can replay a scan from any past date and reach the same answer. The artifact is reproducible, not a one-shot generation.
WARREN produces research — not a trade. The investor sizes the position, decides the entry, and signs the order themselves. The agent carries the cognitive load of the screen and the write-up; the investor owns the decision and the capital.
WARREN is built on the AGENT X pipeline pattern — the same operating shape as Claude Code, with the agent's reasoning streamed to the operator over a live event channel. The pipeline runs the same way every scan; the inputs (universe, prefilter thresholds, model selection) are configurable.
S&P 500 + Nasdaq-100 constituents loaded from the index sources, deduplicated to ~580 unique tickers. Cached for a week to avoid hammering the source. Universe is configurable for private deployments — Russell, FTSE, MSCI sub-sets, or any custom list.
For every ticker in the universe: trailing and forward P/E, P/B, EV/EBIT, FCF yield, debt-to-equity, PEG, market cap, 90-day average volume, distance from 52-week high, dividend yield, current price. Parallel fetch, 12 workers. Snapshot cached for 6 hours so repeat scans don't repeat the slow step.
Before any LLM call: drop the obviously expensive, illiquid, or over-levered names using a configurable rule set. A ticker passes the gate if it satisfies at least N of: P/E < 18, fwd P/E < 16, P/B < 3.0, FCF yield > 5%, EV/EBIT < 14, D/E < 1.5, PEG < 1.5, ≤85% of 52w high, dividend yield > 2%. The defaults are tuned for value-orientation; thresholds are configurable.
Survivors of the gate go to a Claude Haiku ranker in batches of 25. Each ticker gets a numeric score (1–10) on "warren-worthiness" — undervalued, mispriced, structurally cheap — and a one-paragraph reason. Speed: full ranker pass in under 60 seconds on a typical scan.
The top-N (default 10) by ranker score graduate to a Claude Opus analyst. Each name receives a full thesis: bull case, bear case, what would change my mind, catalysts, risks. Output is the analyst-grade research artifact — not a buy/sell signal. Every scan package includes the full Opus deep-dive on the top ten — no per-thesis add-on, no upsell.
Every event — universe loaded, snapshot fetched, ticker passed, ticker scored, thesis written — streams to the operator console over a server-sent event channel. The investor watches the agent reason and can pause, drill in, or kick a re-scan. Live soon at WarrenTrades.com.
The investor opens the console, picks a universe (S&P 500, Nasdaq-100, or both), and triggers a scan. Optionally narrows by sector, market-cap floor, or a custom watchlist. One command — no model selection, no parameter tuning required.
The investor reads the ranked shortlist and the top-ten theses — bull case, bear case, what-would-change-the-call, catalysts, risks. Adds names to a persistent watchlist. Re-asks tomorrow against a fresh snapshot. The agent's reasoning is preserved in the run log — every scan is reproducible, every thesis is replayable.
With the thesis in hand the investor sizes, decides, and signs the position through their broker of choice. WARREN delivers the research artifact — the investor owns the trade. Conviction in, execution out; no black-box signal, no auto-routing, no shared P&L.
WARREN pays off where the investor's edge is decision-making and position sizing, and the bottleneck is generating analyst-grade coverage at scale. Three profiles get the most leverage from the agent — the math compounds fastest when the user is already operating with research discipline.
Self-directed investors who want fundamental research across a wide universe but don't have access to a sell-side desk. WARREN gives them analyst-grade coverage on demand for the price of a scan package — not a sell-side retainer. The live console will be available soon at WarrenTrades.com.
Single-family offices and emerging long-only funds that want repeatable analyst coverage without scaling headcount. WARREN replicates the workflow of a junior research analyst across the full universe and writes the artifact the PM reads. Custom universe and prefilter thresholds available for private deployments.
Systematic teams that already have factor models but want a qualitative narrative layer on top — for portfolio construction, risk overlay, or external communication. WARREN's written thesis is the bridge between a screen output and a position memo. The pipeline is configurable end-to-end.
Live console launching soon at WarrenTrades.com · Private deployments via the engagement desk · Response within two business days