Built to make you an extraordinary quant.

Pyon is the agentic platform for systematic investing. Compose strategies from blocks, let the AI research alongside you, and deploy with the rigor a desk demands.

Pyon - AI Strategy Workspace
Ask the agent to build a strategy →
Total return
+13.9%
Sharpe
1.47
Max drawdown
-12.7%
Win rate
53%
Equity curve · $10,000 start
Passed
90 sessions · daily bars · commission model on89 fills simulated

Everything a desk needs, in one agent

Build, validate, and ship systematic strategies without leaving the canvas.

Ship with confidence

Every strategy is backtested across bull, bear, and chop regimes before it can go live.

Pre-deploy checks4 passed
  • Walk-forward validation
  • Regime sweep across 12y
  • Risk limits within mandate
  • Slippage and cost model
Deploy strategy

Adapts to your mandate

Encode your risk limits and house rules once. The agent enforces them on every build.

mandate.yaml
max_gross_exposure: 1.5x
per_name_cap: 8%
stop_loss: drawdown > 10%

# enforced on every build

- no lookahead in signal windows

- position sizing vol-targeted

- sector neutrality required

The agent reviews your alpha

Pyon flags lookahead bias, overfit parameters, and leverage spikes before you deploy.

pyon agentreviewed just now
strategy/momentum.py
14 -lookback = 5
14 +lookback = 20
15 +# vol-adjusted window

Caught an overfit lookback on the 5-day window.

Describe it. The agent architects it.

Say what you want in plain English. Pyon composes it into modular nodes, triggers, conditions, actions, and AI logic, that compile to a validated, auditable graph. Systems, not scripts.

5 nodes
generated in one prompt
100%
typed and auditable
0 lines
of code to ship
Pyon agentonline
Build a mean-reversion agent on AAPL that buys oversold and stops out past 8% drawdown.

Done. I composed 5 nodes and wired the graph:

Trigger
Market Open
Condition
RSI(14) < 30
Condition
Drawdown < 8%
Action
Buy AAPL · vol-target 10%
Action
Notify desk
Open in builder

Your own models, trained in the Lab

Pick features, train on real market data, validate walk-forward, and drop the signal straight into a strategy as a node. No notebooks, no glue code.

Pyon ML Lab · model-7 (MLP · 3x64)Configuring
Features · AAPL, 5y daily
target: fwd 5d return
Training loss
epoch 0/60 · loss 0.672
Walk-forward validation
2021
53%
2022
56%
2023
54%
2024
58%
2025
57%
Train on the past, test on the next year. Every fold out-of-sample, no lookahead.
Pipeline
Configure features
Train on market data
Walk-forward validate
Publish to builder

AI-powered
agent creation

Describe a thesis. Our co-pilot builds an autonomous agent.

See AI disclosures.

I'd like a strategy that benefits from demand for GPUs.

Here's a strategy that focuses on companies involved in the production and development of GPUs:

40%30%20%10%0%
Jan 01Mar 01May 01Jul 01
AI Strategy+0.0%
Benchmark+0.0%
0.0x
Better returns

Serious backtesting for serious capital

Point-in-time data, regime-aware results, and the risk diagnostics a desk signs off on.

Survivorship-bias-free history

Backtests run on the universe as it actually was, delisted names included, so winners are not quietly assumed.

Universe coveragepoint-in-time
  • Active tickers3,400
  • Delisted included1,210
  • Point-in-time fundamentals12y

Regime-aware results

Performance is broken out across bull, bear, and choppy regimes so a single benign decade cannot flatter a strategy.

Return by regime
  • Bull+24.1%
  • Bear-6.8%
  • Chop+3.2%

Risk diagnostics, not vanity returns

Drawdown, tail risk, and stress metrics sit next to the return, the way an allocator reads a track record.

Diagnostics
Max drawdown
-11.4%
Sharpe
1.83
Sortino
2.41
VaR (95%)
-2.1%

Institutional-grade testing. Not toy backtests.

Live Private Beta

Pyon is Deployed and Operating

First 500 users. Join the beta.

Current Capabilities

Visual builder
Live execution (stocks, crypto, forex)
NL agent generation
Secure auth
Production infrastructure

Expanding

Coming Soon
Enhanced backtesting engine
Collaboration workflows
Institutional integrations