History adjusted for splits.
A split changes the quoted price overnight. Adjusted history keeps the series continuous, so the change is not read as a gain or a loss.
Build strategies.
Advise clients.
Pyon strategy workspace
Pyon Quant tests a trading idea before you trade it.
Describe a strategy in plain English, and Pyon drafts it as a graph you can edit, backtests it on adjusted history and explains the result. Your own AI agent can do the same through the Pyon MCP server.
Describe the rule. See the strategy.
Tell the Pyon agent what you want to test. It drafts the trigger, the condition, both outcomes and the exit as a graph you can edit.
“Buy AAPL when RSI(14) falls below 30. Exit above 70. Define the position size.”
Explore the builder
Illustrative workflow. Review every rule before testing or trading.
Trigger
Market open
Condition
AAPL RSI(14) < 30
If true
Enter with a defined position size
If false
No entry
Exit rule
AAPL RSI(14) > 70
Test the rule. Understand the result.
Backtests run on split-adjusted history, check each rule before a run, and explain what happened, including when a strategy never traded.
Catch the impossible before a run.
RSI(14) > 120
RSI is bounded from 0 to 100. An out-of-range threshold is flagged before it becomes a strategy that never fires.
Model fields and position size are reviewed too.
0 trades can be a useful result.
The backtest can report which entry condition blocked trades and the values it observed in that window.
Illustrative diagnosis, not a backtest result.
Read risk in the right context.
- Strategy backtest
- Drawdown, Sharpe ratio, win rate, and profit factor.
- Portfolio tear sheet
- Sortino ratio, VaR and CVaR, monthly returns, and correlation matrix.
Train a model. Add its signal to a strategy.
Choose market features, train on historical data, and review the signal before using it as a condition.
Choose market inputs.
Five illustrative inputs feed one target, the forward five-day return.
returns_5dRecent price movement
rsi_14Momentum
vol_20Volatility
sma_ratioTrend
volume_zUnusual volume
Five market inputs One illustrative target: forward five-day return
Watch the training curve take shape.
The model fits the selected inputs over 60 illustrative epochs. The curve shows training loss, not returns.
Validate beyond the training window.
Walk-forward checks move through time, with each later window held out from the data used to fit the model.
Fit through 2020-21Test on unseen 2022 data
2022Fit through 2020-22Test on unseen 2023 data
2023Fit through 2020-23Test on unseen 2024 data
2024Fit through 2020-24Test on unseen 2025 data
2025Train on earlier periods Validate on later periods
A learned signal joins the strategy.
The prediction is available as a Strategy Builder condition, ready to review before you run a backtest.
ConditionModel signal · ml_predictReady to review in Strategy Builder
Illustrative walkthrough. Sample values are not a live training run or investment results.
Give your agent the tools to test an idea.
Connect an MCP client to Pyon. Your agent can work through a strategy and return results for you to inspect.
Connect from the client you already use
Codex- Market data
- Strategy building
- Backtesting
- Diagnostics
One request. The tools to work through it.
Your request
Build a momentum basket on AAPL, MSFT and NVDA. Backtest it, then tune it.
- create_strategy
Draft strategy graph returned
- run_backtest
Simulation complete; trades and risk available to inspect
- diagnose_strategy
Entry conditions and execution reviewed
- optimize_strategy
Candidate parameters compared
Review the results in Pyon before deciding what to run.
Connect your agent.
Create an API key under Account in Pyon, then add the MCP server to your client.
One command in your terminal:
claude mcp add pyon -e PYON_API_KEY=pyk_your_key -- npx -y pyon-mcpThis sequence illustrates MCP tool use. It is not a live strategy run.
Pyon Wealth runs each client file from fact-find to review.
Pyon Wealth is for financial advisers. Each client's answers, the figures that follow from them and your advice live in one file, which reports, reviews and Consumer Duty records all draw on.
Every client on one map.
Each client opens as a map of their file. Every card, from reviews and risk to pensions and vulnerability, shows where that part stands and opens the work behind it.
Screens show the Pyon Wealth app with invented clients.
A fact-find that does the sums.
Ask one question at a time or show them all on one page. Answers already on file are filled in, so clients only tell you what has changed.
Clients answer by link or with you.
Send a secure link to answer at home, or fill the fact-find in together during the meeting.
Each answer feeds the file.
Pyon works out figures such as take-home pay, emergency-fund months and time horizon from the answers.
Suggestions wait for you.
Anything Pyon infers stays a suggestion until you confirm it with the client.
Plans and reports from the same file.
Goals are tested against the client's cashflow plan, and the suitability report drafts from the same figures, so nothing is copied across by hand.
Goals against the plan
See whether each goal is met on the median path and on a poor run of markets, and run a cashflow simulation with platform, fund and adviser charges split out.
Suitability reports
Generate a draft from the file, write your rationale for each goal, finalise it with your firm's branding, and send it for the client to read and confirm.
Pensions, tax and Consumer Duty in one place.
The work that often sits in separate spreadsheets lives on the same client file, ready for the next review or file check.
Clients list past employers through a secure link, and Pyon suggests the scheme each one points to. One signature covers a letter of authority for every scheme, which you print, send and track until the provider replies.
The Tax Centre lists unused allowances across the book, such as ISA, pension annual allowance and the CGT exemption. Each decision to act, wait or dismiss is logged on the client's history.
Record vulnerability against the FCA's four drivers, with the client's consent, the support you will give and a date to look again.
Log each complaint with its evidence. The register works out the eight-week response date and flags any complaint that passes it.
Research and market notes.
All articlesLoading articles…