It scans the market every half hour, reads filings, insider buys and news, decides with a written thesis, and places every order inside hard limits you set. Then it reviews its own results each night and rewrites its playbook.
Watches the marketEvery 30 minutes in market hours
Explains itselfThesis and confidence on every trade
Hard limitsStops, caps and a kill switch in code
Learns nightlyRewrites its playbook from results
Paper firstLive only when it has earned it
Your phoneStart, stop, halt from anywhere
Home
paper…
Portfolio value—
Today—
All time—
vs SPY—since tracking began
Drawdown—
Positions—
AI cost—
Equity curve
What the AI is doing
—
…
—
Open positions
Crypto sleeve
Today's plan
—
Is the rule worth it?
Backtest of the exact live rule on Alpaca's daily history, after fees and slippage.
AI research bookshadow · no orders
Claude researches the market every day and picks up to 6 of Alpaca's 32 coins on its own. Its picks are tracked at real closes, after fees and spread, against the rule and against holding BTC. It places real paper orders only if it wins after a fair run.
Live
connecting…
·
Prices tick every 5 seconds. Decisions, orders and agent log lines appear here the moment they happen.
Connected services
OKAY DESKreactortapebooklab
--:--:--UTC · live
LIVE WIRE
The reactor · agents ride the ring · one ticket at a time · the book in the core
Injector banksix live series
—equity
Control stackticket · stages · flow
Book P&Lone book · balance history · after costs
—your money · equity now
—P&L since start
The tape
—vwap · 20-bar bands · volume · live order book
The traceevery step the desk takes
The pricerbar returns · where the last one sits
Signal spectrogramsix factors · top 24 names
every number from the journal or the broker · the motion is the data moving · nothing illustrative
connecting
UTC --:--:--next stock cycle —crypto decides in —
Pipeline · what the system is doing right now
The team · every agent, what it is doing, and what they tell each other
room · hand-offs between agents
The lab · one change a night, tested by code against the live strategy
Account vs SPY and BTC
Risk limitsenforced in code
Book · click a row for the full reason
Tape · livedecisions, orders, agent log
Reasoning
Costswhat it really costs
every number from the journal or the broker · nothing illustrative
Chat
tell it to trade, control the agent, or teach it
What you've taught it
Everything here is injected into the agent's prompt on every cycle, above its own playbook and below the code-enforced limits.
Activity
every cycle, including what it rejected and why
Positions
Closed trades
Symbol
Exit
Return
P&L
Why closed
Brain
what it believes, what pays, what it learned
Scorecard by signal
realized, last 100 trades
Signal
Trades
Win
Avg
PF
Factor scoring model
#
Symbol
Score
Mom 12-1
Trend
52w
Lessons it wrote itself
Playbook
Operator instructions
from Chat
Learned tunables
Controls
Agent
Halt keeps positions and their stops but blocks new buys. Kill closes every position at market, stops the agent, and stays locked until you resume.
Settings
Sliders can only tighten the limits in config.yaml, never loosen them.
Account
Mode: . Live is switched only by an admin from the master console.
Hard limits
code-enforced
Log
Balances
Keys & API
this account only · stored encrypted
Broker & model keys
Blank fields keep their current value. Keys are write-only.
Passkeys
Face ID · Touch ID · security key
API token
Lets another program or AI agent read this account, place risk-checked orders, teach the agent and talk to the assistant. Scoped to this account. Shown once.
Connect an AI (MCP)
Claude Desktop, Claude Code or any MCP client. Tools: overview, positions, activity, scorecard, place_order, close_position, agent, teach, chat.