Back to AI Research Agent

Gradient Deep Research · Global Agent Network

A research desk as wide as the market.

Ask it anything — a market-wide screen, a pointed question, a single ticker. Autonomous agents read filings, fundamentals and news across thousands of global equities, around the clock — roughly 500× the coverage of a human analyst, at the depth of one. Every answer is researched against a point-in-time database, then turned into forecasts and clear recommendations. The breadth is the edge: Sharpe scales with the square root of independent positions.

15,000+ equities under coverage 24/5 continuous research Zero look-ahead · point-in-time data Forecasts · clear recommendations
gradient://network · agent activity monitor — UTC
DAY / NIGHT · LIVE — AGENTS DO NOT KEEP MARKET HOURS
Agents active now
Deep dives · last 24h
Evidence queries today
M
Scenario paths today

Live network view. Activity figures shown are illustrative.

Why breadth wins

Sharpe scales with the square root of breadth.

IR ≈ IC × √N  —  skill × √breadth

You just watched thousands of agents work a single market in parallel. That width — not any one brilliant call — is the edge, and a law says why. The fundamental law of active management (Grinold, 1989): risk-adjusted return is skill times the square root of the number of independent positions. A human desk can only add names by diluting depth. Our agents hold depth constant and scale the count — the same per-name skill applied to 100× the breadth compounds into 10× the Sharpe ratio.

Sharpe multiple vs. independent positions

EQUAL SKILL PER NAME (IC HELD CONSTANT) · INDEXED TO A 30-NAME DESK · LOG SCALE

VIEW AS TABLE
Independent positionsSharpe multiple
100.6×
30 — human desk1.0×
1001.8×
3003.2×
1,0005.8×
3,000 — Gradient fleet10.0×
~30 names
What one excellent human analyst can cover with genuine depth.
~3,000
Independent positions the book runs at once — the residual bets that remain from the ~15,000-name coverage universe once common factors are neutralized.
10×
Sharpe multiple at equal skill per name. Breadth is the only free lunch left, and it doesn't scale with headcount.

Breadth only counts when the positions are genuinely independent — so the common factors (market, sector, size) are stripped out first, leaving each bet to stand on its own residual. Illustrative relationship; assumes constant information coefficient and faithful transfer of forecasts into positions.

Agent output · verbatim

Watch one agent reason.

Breadth is only worth 10× if every one of those thousands of positions is real research — not a guess. So here is the reasoning a forecast and a recommendation are built on: every finding runs the same fixed contract, traceable end to end, with absolute neutrality. Step through three from one cluster run:

agent://cluster-run · agrochemicals
    earnings_swing_vs_special_charges 5 evidence queries · 1 self-correction · audit trail complete

    See it on your own names

    Point the fleet at a company you know.

    The fastest way to trust a research desk is to test it where you already have a view. Screen the whole market, drill into a single name, or let it monitor your book overnight — early access runs the same agents you just watched, on the names that matter to you.

    gradient research 2330 ↵ run
    01 · ASK

    Ask it anything

    A ticker, a market-wide screen or a pointed question — across 15,000+ covered equities.

    02 · WORK

    The fleet researches it

    An agent forms a hypothesis, interrogates a point-in-time database and self-corrects — zero look-ahead, absolute neutrality.

    03 · ANSWER FORECAST + RECOMMENDATION

    evidence → a forecast with a long-term value → a clear, conviction-scored call · full audit trail

    Then it keeps watching: every night the fleet re-runs your book and turns new evidence straight into an updated recommendation.

    Read it against what you already believe. That's the test.

    Gradient Deep Research · Early Access

    Machine-scale research.
    Human-grade depth.

    Test it on the names you know best — and judge the results for yourself.

    The company examples and figures on this page are illustrative — they show how Gradient Deep Research works (its research workflow, evidence trail and output format), not actual holdings, forecasts, or investment recommendations.