Crestant
Not a broker. Not a screener. An agent.

An investor
who never stops
learning you.

Crestant reads the market around the clock, writes theses you can audit, proposes trades with invalidation rules, and reflects after every win and loss — so the advisor you have next year is sharper than the one you have today.

Paper-traded since March 2024
Theses written 12,481
Avg citations 6.2 / thesis
S&P 5005,847.12+0.42% · US 10Y4.18%−0.03 · DXY103.24−0.21 · WTI$79.14+1.12% · VIX14.82−2.40% · NVDX$412.08+2.31% · ORIN$84.22−0.87% · TDWN$27.91+4.12% · S&P 5005,847.12+0.42% · US 10Y4.18%−0.03 · DXY103.24−0.21 · WTI$79.14+1.12% · VIX14.82−2.40% · NVDX$412.08+2.31% · ORIN$84.22−0.87% · TDWN$27.91+4.12% · S&P 5005,847.12+0.42% · US 10Y4.18%−0.03 · DXY103.24−0.21 · WTI$79.14+1.12% · VIX14.82−2.40%
How it thinks

A portfolio manager's process, written into software.

Most investing tools stop at data. Crestant treats data as input to a reasoning loop: read, reason, propose, reflect — the same loop a good human PM runs, except it runs while you sleep and remembers everything.

08:42 scan · 847 tickers
08:43 read · fed speakers calendar
08:44 read · SEC filings 13F delta
08:45 read · supply-chain bulletins
08:46 read · social sentiment Δ
08:47 correlate · semi capex + TSMC util
01 · READ

Reads the market continuously, not on your command.

The agent scans macro, sector, supply chain, and filings on a cadence it sets for itself — every few minutes during opens, hourly overnight. You don't ask. You get a morning brief.

"Grid modernization thesis holds if all three are true: DOE tranche lands on schedule, regulated returns don't compress, and capex flows into transmission not generation."
DOE Capex mix ROE floor
02 · REASON

Writes a thesis before it writes a ticker.

Every potential trade starts as a short written thesis: the observation, the supporting pillars, the time horizon, and explicit invalidation criteria. No thesis, no trade.

proposal 74% conf
Rotate FLTA → ORIN · $2,695 net
Approve Discuss Reject
stop −7% · tp +14%
03 · PROPOSE

Proposes. Always proposes. Never surprises you.

Trades arrive as proposals — sized, justified, with the counter-argument written next to the argument. You approve, redline, or refuse. Nothing executes silently.

Journal · Apr 16
"I'm slow to adjust when China demand data shifts. Last 6 months: 4 of 5 misses. Adjusted the model weighting by +18%."
weight revised · 4 signals impacted
04 · REFLECT

Keeps a journal. Re-reads it. Gets better.

After every position closes — winner or loser — it writes a reflection. Patterns compound: what it got right, where it was early, when your judgment improved its own.

Guardrails

An agent with judgment is only useful if it also has limits.

Read-only by default

Order execution stays off until you enable it. Even then, every trade needs your approval unless you explicitly delegate.

Style guardrails

Your investment style sets hard limits on position size, sleeve allocation, and drawdown. The agent can't override them. Only you can.

Full audit trail

Every thought, citation, and reasoning step is logged. Click any trade to see exactly what the agent was reading when it decided.

Invalidation first

No position enters without written invalidation criteria. The agent exits on rule, not on feeling.

Your data, your model

Your trade history trains your agent only. It never informs another user's model.

Cool-down windows

After losses, the agent enters a reflective pause before proposing similar trades. No revenge trading.

Track record

Honest numbers, published monthly.

We publish each style's paper-traded performance, including the bad months. Losers are annotated with the reflection entry that followed.

Read the methodology
Balanced style · since Mar 2024
+26.4% vs SPX +17.8%
1M 3M 1Y All
Win rate
63%
Avg hold
24d
Max DD
−6.2%
Sharpe
1.42

Your agent starts learning the moment you sign up.

Start in paper-trade mode. Approve a few trades. Decide together when to go live. No lock-in, no minimums.