DEXLESSAlphaDesk Autopilot 中文 Enter the cockpit

POST-TRADE FUND OPERATIONS · AI COPILOT · BUILT FOR SUPERVISION

Give it a goal.
It plans first, works step by step,
and waits for you at every decision.

Reconciliation, P&L explanation, mandate checks, rebalancing proposals — the copilot does them on your real trades and positions. Every thought, tool call and output is visible. Approve, reject and change autonomy are always human clicks, and every action lands on a hash chain.

Enter the cockpit See it work No stock picking · No discretionary trading · Licence and discretion stay with you
Copilot studioillustration · fictional data
Do the books tie out? Run a full reconciliation on the latest book date, list every position that doesn't match, then give the investment committee a three-sentence conclusion.
Plan0/5
Confirm fund state, autonomy level and data-quality flags
Run a full reconciliation on the latest book date
Attribute the residual by position, trace to trades [T#] and price sources [S#]
Verify the decision ledger; single-day or persistent break?
Write the three-sentence committee conclusion
The copilot asks for your approval
Approve P002 · trim stock A to 22% weight
All five pre-compliance checks pass; also clears the theme-exposure breach.
ApproveReject
Latest outputReconciliationResidual traceLedger

Break radar

book date · 19/26 checks pass · book vs market
Stock A · price source+109.8%
Stock B · price source+55.4%
Stock C · price source-41.5%
Avg cost · rebuilt from trades0 diff

Daily P&L decomposition

price +48,922 · FX -8,518 · residual -9,870

Conclusion & recommendation (for the IC)

written by the copilot · ledger verified
Residual
+55,357 (8.4× tolerance)
Root cause
Book valuation source mis-set, not a missing trade
Advice
Ask the custodian to resend three days of closes; no stop-loss until cost basis is confirmed
Every number comes from the engine; the copilot only looks up, traces and writes.

Post-trade operations, handled by a copilot that thinks — inside boundaries

You are not short of pre-trade research. What nobody does for you is month-end close, the committee asking "why did we lose", and the auditor asking "did AI take part, and who approved". The copilot takes over everything after the investment decision — tidying, explaining, checking, recording. Humans only decide.

Plan first, then execute

Every goal starts with a 3–7 step plan, executed with tools one step at a time, each with a one-line result. You see what it thinks and what it looked up.

Numbers from the engine, words from the model

Valuation, P&L decomposition, reconciliation and mandate checks are deterministic code; every conclusion cites trades [T#] and data sources [S#].

Stops wherever a decision is needed

Approve, reject, change the autonomy level, emergency stop — an approval card appears and waits for your click. A rejection is respected and recorded.

Every action on a hash chain

Proposals, decisions, executions, overrides and level changes are chained with SHA-256; export an audit pack in one click, including a draft AI-use disclosure.

AUTONOMY LEVEL A dial per workflow. Your governance committee decides how far to turn it; L4 is locked. 


  






  

    
The four questions every investment committee asks

    
Each answer is a number computed by code, with trade and market-data provenance; the copilot only looks up, traces and writes it in plain language.

  

  

    

      
Why did we make or lose money this month?

      
Daily P&L split into price, FX, dividends, fees, flows and residual; a residual beyond tolerance is traced to the position and to the trade.

      
6 bucketsP&L decomposition · residual traced to [T#]

    

    

      
Do the books tie out?

      
Cost, quantity and realised P&L are rebuilt from the trade flow and checked line by line against your book; a book close that disagrees with the market is flagged with the gap.

      
Line by linebreak radar: position × check

    

    

      
What should we adjust?

      
Proposals only when the mandate is breached; each one simulates the post-trade world on paper, re-runs every limit, and lists the rejected alternatives with reasons.

      
Pre-checkedcompliance checks, alternatives, signature

    

    

      
Who approved it?

      
Proposal, approval, rejection, execution, level change, emergency stop — all on one SHA-256 chain; alter one entry and the chain breaks.

      
One chainaudit pack in one click

    

  






  

    
The unit of trust is the decision record, not the model

    
Every action — proposed by the agent, clicked by a human, or written by the copilot — is one entry on the same append-only chain, and each entry's hash includes the previous one. This is what it looks like (illustration).

  

  

    
#15D+42executionP002 executed on paper after approval · trim 451 sharesb4ad9b01ca

    
#14D+42decisionP002 approved · IC agreed (weight 40% → 22%, 5/5 checks pass)3a13951e4a

    
#13D+3reconciliation26 checks · 7 flagged · residual +55,3572003a688f8

    
#12D+2data gateP003 cost basis 116% off the trade-day close → confirm first, no trade proposed14248e2bdc

    
#11D+0proposalP001 trim to 22% target · checks 5/5 · 2 alternatives rejected52570d2f17

    
#10D+0levelL4 blocked: high-risk decisions may not be fully automated; governance approval required7e14366e2b

  

  
Chain verified · each entry = sha256(previous hash + payload)Tamper with any entry and everything after it breaks






  

    
Three books, one set of market prices

    

      The actual book is your real ledger. Your book is the actual book plus the paper actions you approved. The shadow book is what would have happened if you had taken every agent proposal.
      Where the lines diverge is the trace of human overrides — regulators ask you to track it; almost no vendor does.
    

    

      
Actual (marked to market)

      
With your approved actions

      
If every proposal were taken

    

  

  





  

    
DEXLESS RISK ENGINE

    
One risk discipline, across financial markets

    
Markets change; the discipline doesn't: build a self-baseline for every account → trigger on breaches → intercept before execution → record every action. The engine is built by Dexless and lands as a different product per segment.

  

  
  

    

      
CRYPTOLIVE

      
Dexless AI Brain

      
The harshest segment — 24/7, extreme volatility and leverage. Behavioural forensics, self-baseline triggers, pre-trade interception: the engine was hardened here first.

      
    

    

      
FUNDS & ASSET MANAGEMENTTHIS PRODUCT

      
Dexless AlphaDesk

      
The regulated segment. The same engine, plus mandate-as-code, approval cards, the decision ledger and audit packs — in the language of the FSC, MAS and FSB.

    

    

      
NEXTPLANNED

      
Bank prop desks · family offices

      
Same engine, different books. Segments will grow; the discipline won't change.

    

  

  

    
Why numbers come from the engine, never from a language model — we measured it

    

      
100% vs 25%behavioural-anomaly precision
engine vs general LLM (GPT-4o)

      
30ms vs 2,262msper-analysis speed (75×)

      
3,705real perpetual-futures trades tested; the LLM raised 12 false positives and missed one class entirely

      
$0engine inference cost — the reason Dexless AlphaDesk computes with code and lets the model only explain

    

    
Internal benchmark, Mar 2026 · ground truth = deterministic pipeline · GPT-4o, temperature 0, JSON mode · full method and per-class numbers available in the deck

  

  
In crypto, the engine learned discipline where there is no referee. In funds, it submits that discipline to one.






  

    
Three things we don't do

  

  





  
TRY IT NOW

  
Give the copilot a goal and watch it work.

  
The studio ships with a demo book. Type one line — "Do the books tie out?" "What should we adjust?" — and watch it plan, look things up step by step, and stop for your decision. Bring your own books if you like (account and holder fields never enter the system).

  Enter the cockpit
  
Asset managers · private funds · family offices · bank prop desks · starting in Taiwan, architecture aligned with FSB / MAS SAFR / FSC AI guidance