Judgment
What it decides, and what it refuses.
A case is put against every candidate before capital moves, and what it decides is ranked against everything else it could be doing instead.
AI Agentic Trading
A trading desk that
thinks for itself.
An AI agentic trading desk for macOS and Windows. Multiple strategies compete for one pool of capital, under pre-trade risk checks and broker-resident protection.
Pre-trade risk
Broker-resident stops
Execution analytics
Post-trade reconciliation
SIP and OPRA market data
Judgment
A prosecutor stands between
the green light and your capital.
Clearing the entry rules is not permission to trade. It is permission to be argued with. Every surviving candidate has a case built against it from evidence the desk already holds, and only those that survive it are funded.
Every refusal is recorded and scored later against what the market actually did, so the cost of saying no is a measured quantity rather than an article of faith.
Its accountability is structural, not aspirational.
Every refusal is carried through the same lifecycle as a funded trade, so what it saved or cost becomes a measured quantity rather than a claim. That record is reported, and if it turns against the desk, the weight given to those refusals steps back until the evidence restores it.
Nothing here waits on a human.
There is no approval queue, no discretionary override and no setting that lets an operator wave a trade past the chain. You set the account and the risk limits. It runs the desk, whether you are watching or asleep.
Entry gauntletIdle
Candidate forms·
Worth the cost·
Conditions allow it·
Data can be trusted·
Survives the case against it·
Within its limits·
Sized and bounded·
Protection rests at the broker·
StandbyEvery candidate walks this path. The ones that clear it reach your account with a stop already resting at the broker, and every decision, either way, is recorded and later graded.
Schematic of the decision path. Not a live feed.
Intelligence
It knows
how much it knows.
Almost no trading system can tell you whether it is getting better, because nothing in it is keeping score of itself. This one carries a standing measure of its own competence, assembled from four components, and it keeps the history so that improvement is something you can see rather than something you argue about.
It is assembled from several measures of the desk’s own record, each computed from the same tables it trades from, and the history is kept so that improvement is something you can see rather than something you argue about.
Every component is computed from the same tables the engine trades from. Nothing in it is a matter of opinion, nothing responds to how the last few days felt, and nothing moves the measure except outcomes that have actually been recorded.
The agent loop
Every trade earns its way
to your account.
This is what agentic actually means here: the software perceives, judges, acts, grades its own result and adjusts, without a human in the loop. No discretionary override, and no trade that skips a step. A chain of independent checks stands between a signal and your capital, each one recorded as it happens. Underneath runs a second loop, which measures how well those checks are performing and tunes them from what it finds.
- 01Observe What the desk can see across every market it follows.
- 02Judge A case is put against the trade before any capital moves.
- 03Act Entry and its protection go out together wherever the venue allows it, and a position that cannot be protected is closed.
- 04Grade Every decision is scored against what the market then did.
The fourth step is the one most systems never take. Grading feeds back into the first, so the desk that trades tomorrow is not quite the desk that traded today. Almost every self improving system learns only from what it took. This one also learns from what it declined, which is the great majority of what a disciplined desk does.
Execution
The edge has to survive
the fill.
Expected edge is a claim. What the account receives after spread, fees and slippage is the result. The desk measures that gap continuously, from its own execution record, and prices it back into the next decision rather than discovering it at month end.
- What execution actually costs
- The desk keeps its own record of what its orders actually cost, rather than relying on an assumption inside a cost model. That record is built from what this account saw, not from a vendor’s average, and where the data is conservative it is left conservative.
- Priced in, not reviewed later
- What execution actually costs is priced into later decisions automatically. Execution quality is an input the desk acts on, not a number somebody reviews after the quarter closes.
- Sizes that can always be protected
- Order sizes are computed so that a protective stop can always be placed against them. A position that cannot be protected is not a rounding error, it is an unbounded one, so the desk closes it rather than holding it.
- Books that match the brokerage
- Realised profit, fees and history are rebuilt from the broker’s own record rather than from what the software intended to do. Anything that happens while it is not running, an overnight stop or a manual close, still lands in the books, because history that must match the brokerage has to come from the brokerage.
- Pre-trade account checks
- Account state is verified before the engine may trade and re-checked on a cycle while it runs: status, restrictions, and the regulatory minimum equity for margin under FINRA Rule 4210(b)(4). The guard either passes or it stops the engine and names the reason. There is no approval step in it.
How it works
Three things it does.
All of them measured.
They divide into what the desk decides, what it remembers, and what it changes about itself. Each is scored by the machinery it supervises, and each earns its authority from its own record.
Specific thresholds, weights and formulas are withheld throughout this page. Licensees receive them in full.
Memory
What it keeps, so that it can learn at all.
Every yes and every no it has ever made, kept and later scored against what the market actually did. Exits are scored as carefully as entries.
Learning
What it changes about itself, and on what evidence.
Capital moves toward what its own record says is working. It re-studies itself after every close, and a second layer reviews the whole thing on a cadence and reports what it finds.
Bedrock
Seven things no intelligence
in this system may touch.
Under every autonomy mode, including full automatic. A finding that would require weakening any of these is skipped and explained rather than applied, and the list itself is one of the protected items, so nothing in the system can quietly shorten it.
- Protective stops
- They rest at the broker, never only inside this process. If the engine dies mid position, a stop already resting still fires. Equities enter as bracket orders so no window exists where the position is unprotected. Crypto gets a resting stop immediately after the fill, and if that stop cannot be placed the position is closed rather than held naked.
- The kill switch
- In the menu bar and on the dashboard. Cancels every order, closes every position. Nothing can gate it, defer it, or reason with it.
- The daily loss stop
- Trading halts at the limit you set for the day. Not a suggestion, not a soft warning, and not a control any AI layer may widen.
- The profit target rule
- A stopping rule, deliberately. Reaching the target ends the day. It never authorises larger size to chase more, and size never increases in response to a setback. Martingale logic fails the charter on sight, and there is none anywhere in this system.
- Pre flight refusal
- If the self test fails, trading is blocked and the app names the failing step. There is no override.
- The sizing ramp and caps
- New order routing code proves itself at reduced size before it trades at full size, and concentration caps bound how much of the account any single idea may become. Both are automatic and require nothing from the operator.
- The implausible order guard
- The last line between a bad number and a real order. An order that makes no sense against account state never reaches the broker.
Pre flight
Before it risks anything,
it risks a dollar.
The engine will not trade until an automatic self test passes. Nobody is asked to approve it. It simply runs, and takes about two minutes. The fourth step is the one that matters, because it proves stop orders genuinely reach the broker with real money before anything larger is at stake.
- Verify credentials, and confirm the account is active and unrestricted
- Confirm market data is genuinely streaming for each asset class
- Check the trading clock
- Place a real order of about a dollar in risk, confirm a protective stop is resting at the broker, close the position, and verify the fees booked correctly
- Reconcile the local database against real broker state
Then the ramp. Early trades run at a fraction of full size, stepping up only once the live order path has proven itself over a run of real fills.
The charter
A change that adds expected profit while weakening a loss bound fails this charter.
Governing document. Loaded verbatim into every audit the machine performs on itself.
Every judgment the system makes, whether a control change, a veto, an audit finding or a line of code it proposes, is tested against two hard goals and four questions. Which goal does this serve? What graded evidence supports it? How will success be measured? What is the damage if it is wrong, and what bounds that damage? A change that cannot answer all four is an observation, not an action.
The honesty requirements are enforced structurally rather than aspirationally. Claims are graded, and the verdicts outrank the rationale that produced them. Changes carry receipts: what changed, why, the value that was persisted, and the metric that will later prove or disprove it. Every part of the system is held to the standard it applies to others.
Perception
Most of what a chart knows
is already in the price.
So the desk draws on sources that are not, and on shapes a chart cannot show you. Filings, disclosures, funding, calendars, the book itself, and the changing geometry of the market as a whole. Every feed is best effort by design: a dead feed gates nothing, and the engine keeps flying.
- Congressional trades
- Insider filings
- Activist stakes
- Funding
- Earnings calendar
- Volatility
- Market microstructure
- Cross asset shape
The application
A native desktop app.
Not a dashboard in a browser tab.
A native app, resident in the menu bar on macOS and on the desktop on Windows, watching an engine that runs as its own supervised process. Both run on your machine and talk to each other locally. The engine reaches out to your broker and its data sources, and nothing accepts inbound connections.
Set two things, then leave it
Broker keys and risk limits. The keys go into the operating system credential store, the macOS Keychain or the Windows Credential Manager, and are never written to disk. After that the engine runs its pre flight and starts. On the next launch it resumes in whatever mode it last ran, and after a reboot the login item brings it back without anyone pressing anything.
It shows its reasoning, not just its results
Every row in the ledger is inspectable: what the desk saw, what it decided, and, days later, what the market did about that decision. The dashboard carries live market context derived from the same data the engine trades on, so the screen can never disagree with the trading logic.
Watch its mind change
The learning feed records every adjustment the desk makes to itself, with the evidence that caused it. Benches, revivals, raised gates, refreshed priors. It is the difference between a system that claims to adapt and one that shows you the receipt.
Bring your own signals
Alerts from charting platforms, or anything that can post JSON, enter as ideas and never as orders. They clear every check anything built in has to clear, and they are treated as one more strategy with no head start, set aside if they do not pay. An alert carrying a quantity is refused outright, because sizing is never the sender’s job.
The suite that guards all of this includes a check which drives a live regulatory filing all the way to a protected order at the broker, with no human step anywhere in the path.
Tech Specs
What you take delivery of.
- Platform
- macOS 13 or later on Apple silicon and Intel · Windows 10 or later
- Engine
- Python, one supervised process
- Interface
- Native on each platform. SwiftUI and menu bar resident on macOS, a native desktop app on Windows
- Brokerage connectivity
- Connects to your own brokerage account over its API · additional broker adapters in migration
- Equities
- US listed common stock and ETFs, across every venue on the consolidated tape, in cash or margin accounts
- Options
- US listed options priced from OPRA. Single leg, verticals, and multi leg defined risk spreads, sized so the worst case is fixed at entry
- Crypto
- Traded around the clock, including the hours and weekends when equity markets are closed
- Market data
- SIP consolidated tape for equities, OPRA for options, gated on verified entitlements. Degrades to single-venue quotes and says so on screen
- Strategies
- Multiple, competing for a single pool of capital
- Allocation
- Evidence weighted across strategies, adjusted for cost
- Order protection
- Broker resident. Bracket for equities, resting stop for crypto, defined risk for spreads
- Order arithmetic
- Sizes computed so a protective stop can always be placed against them
- Pre-trade risk
- Account status, restrictions and FINRA Rule 4210(b)(4) margin minimum, checked before trading and on a cycle thereafter
- Execution analytics
- The desk keeps its own record of what its orders cost, and prices it into later decisions
- Reconciliation
- Realised P&L and fees rebuilt from the broker’s own record. Positions and protection reconciled at start up before any new order
- Research stack
- Backtester with a real cost model, options repricing, and validation built to reject findings that do not hold out of sample
- Prior depth
- Equity index from 1993. Bitcoin from 2014
- Storage
- Local database and research store. Nothing leaves the machine
- Credentials
- Held by the operating system. macOS Keychain, or Windows Credential Manager. Never written to disk
- Verification
- Automated test suites. Backtests and a full year single account simulation reproducible from one command
- Deployment
- One machine, starting automatically at login. Resumes trading after a reboot unattended
- Documentation
- Owner’s guide, field manual, data flow reference and the governing charter
Licensing
Three ways in.
Sold to operators and firms who intend to run it, not to redistribute it. Every tier includes the charter, the owner’s guide and the reproducible research, because a system you cannot audit is one you should not run.
Evaluation
Fixed term, paper account only
- The complete application, restricted to a paper account
- Full pre flight, dashboards and the full decision history
- Run the backtests and the year simulation yourself
- Owner’s guide and charter
Operator
Single operator, cleared for live
- Everything in Evaluation, cleared for live trading
- Signed application, with updates for the license term
- Your own broker credentials and your own account, always
- Private support channel with the engineering team
- Onboarding covering the risk settings that actually matter
Source
Perpetual, with build rights
- Full engine and application source, with build rights
- The research stack, the priors and the study code
- One broker adapter port included
- White label naming and icon
- Architecture handover with the engineering team
Inquiries
Tell us what you
would run it on.
Account size, the asset classes you care about, and whether you intend to operate it or build on it. Terms follow from that.
The thresholds behind all of this are withheld for the same reason the desk refuses most of its own signals: the value is in the calibration, and calibration given away is calibration destroyed. Licensees receive it in full.
Answered by an engineer, not a queue
Questions
What agentic trading
actually means here.
- What is AI agentic trading?
- An agentic system does not wait to be told what to do at each step. It perceives the market, forms a judgment, acts on it, measures what happened, and adjusts its own behaviour from the result. TradeAgentic.ai runs that whole loop continuously across crypto, equities and options, with no human approving individual trades.
- Does it really trade without approval?
- Yes. There is no approval queue and no discretionary override. What you control are the boundaries: which account it uses, how much it may risk, and when it must stop for the day. Inside those boundaries it decides on its own, and every decision it makes, including every refusal, is recorded and later graded against what the market did.
- How is this different from a trading bot?
- A bot executes rules you give it and cannot tell whether those rules are still working. This desk argues against its own candidates before committing, scores the value of its own refusals, withdraws capital from strategies its evidence says have stopped paying, and re-studies its priors after every close. The rules are not the product. The judgment about the rules is.
- Which broker does it connect to?
- It connects to your own brokerage account over that broker's API, using your own credentials, which are held in the operating system credential store and never written to disk. Supported brokers and the market data feeds available through them are confirmed during licensing. Additional broker adapters are in migration, and a source license includes one broker adapter port.
- What happens if the software stops while a position is open?
- Protective stops rest at the broker, never only inside the process, so a stop already placed still fires if the software is not running. On the next start it compares its records against the broker, adopts anything it does not recognise, and re-arms any missing stop before it is allowed to trade again.
- Can I see why it did something?
- Every decision it has made is inspectable: what the desk saw, what it decided, and, days later, what the market did about that decision. The learning feed separately records every adjustment the desk has made to itself, with the evidence that caused it.
- Do I need to write code or pick strategies?
- No. There are no strategy switches in the product. Multiple strategies compete for one pool of capital, and the desk decides between them from their measured results. You set two things: your keys and your risk limits.
- What does it run on?
- One machine running macOS 13 or later on Apple silicon or Intel, or Windows 10 or later. It starts automatically at login and resumes trading after a reboot without anyone pressing anything. No market or account data leaves the machine.