Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.quanux.org/llms.txt

Use this file to discover all available pages before exploring further.

QuanuX strategies are built from modular components. You start in Python — where iteration is fast and the data science ecosystem is available — then graduate the logic to compiled code once you’re confident in the approach. This page covers how to write a strategy from scratch and how to use the Foundry API to generate higher-performance equivalents.

​
Strategy architecture

Every strategy in server/strategies/ is assembled from four component types, each inheriting from StrategyComponent defined in server/strategies/base.py.
ComponentBase classResponsibility
SignalSignalModuleAnalyze market data; return LONG, SHORT, or NEUTRAL
EntryEntryModuleGenerate an order dict when a signal is present
RiskRiskModuleCalculate stop loss, take profit, and trailing stop levels
ExitExitModuleDecide whether an open position should close
A CompositeStrategy wires these components together and drives the on_bar loop.

​
Writing a strategy in Python

The example below is drawn from the SmaCrossover strategy in server/strategies/full/SmaCrossover/. 
from server.strategies.base import SignalModule, SignalType, BaseParameters
from typing import Type
import pandas as pd

class SmaSignal(SignalModule):
    """Simple Moving Average Crossover Signal.
    Long when Fast SMA > Slow SMA. Short when Fast SMA < Slow SMA.
    """
    def define_parameters(self) -> Type[BaseParameters]:
        class Params(BaseParameters):
            fast_period: int = 10
            slow_period: int = 20
        return Params

    def on_bar(self, bar_data: dict) -> SignalType:
        if not hasattr(self, '_history'):
            self._history = []

        close = bar_data.get('close')
        if close is None:
            return SignalType.NEUTRAL

        self._history.append(close)
        slow = self.params['slow_period']
        if len(self._history) > slow + 20:
            self._history.pop(0)
        if len(self._history) < slow:
            return SignalType.NEUTRAL

        series = pd.Series(self._history)
        fast_ma = series.rolling(window=self.params['fast_period']).mean().iloc[-1]
        slow_ma = series.rolling(window=self.params['slow_period']).mean().iloc[-1]

        if fast_ma > slow_ma:
            return SignalType.LONG
        elif fast_ma < slow_ma:
            return SignalType.SHORT
        return SignalType.NEUTRAL
Define parameters using a nested Pydantic model inside define_parameters. QuanuX validates and coerces all values at runtime when you call update_parameters.

​
Using the Foundry to generate code

Once your Python prototype is working, submit it to the AI Foundry to generate a faster equivalent. The Foundry dispatches your request as a NATS payload and returns a job ID immediately.

​
POST /api/foundry/forge

Submit a generation job for a strategy component. Request body
​
component_type
string
required
The type of component to generate. Accepted values: indicator, entry, exit, strategy.
​
name
string
required
The name for the generated component, used as the file and class identifier.
​
target_lang
string
required
The output language. Accepted values: python, cython, cpp.
​
target_version
string
Semantic version string for the generated artifact, e.g. 1.0.0. Optional; defaults to null.
​
prompt
string
Natural-language description of the strategy logic. The Foundry uses this alongside the component type and name to guide code generation.
Response
​
status
string
required
Always "accepted" on success. The job runs asynchronously.
​
job_id
string
required
Unique job identifier in the format job_xxxxxxxx. Use this to correlate NATS telemetry events back to your request.
Example: generate a Cython signal component 
curl -X POST http://localhost:8080/api/foundry/forge \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "component_type": "strategy",
    "name": "SmaCrossover",
    "target_lang": "cython",
    "target_version": "1.0.0",
    "prompt": "SMA crossover: long when 10-period SMA crosses above 20-period SMA, short on the reverse"
  }'
Response
{
  "status": "accepted",
  "job_id": "job_3f8a1c2d"
}

​
POST /api/foundry/verify

After the Foundry generates code, run equivalence testing to confirm the generated version produces mathematically identical output to your Python prototype. Request body
​
strategy_name
string
required
The name of the strategy to verify. Must match the name used in the forge request.
Response
​
status
string
required
Always "accepted". Verification runs asynchronously in a deterministic sandbox.
​
job_id
string
required
Unique job identifier for this verification run.
Example
cURL
curl -X POST http://localhost:8080/api/foundry/verify \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"strategy_name": "SmaCrossover"}'
Response
{
  "status": "accepted",
  "job_id": "job_9b4e7a01"
}

​
Git-as-Governance

Before a generated strategy can be promoted to C++ and deployed to the live spreader, it must be committed to your QuanuX repository. The spreader verifies the SHA-256 of every strategy artifact against a signed commit. An uncommitted or unsigned strategy will be rejected at deploy time.
Do not modify generated Cython .pyx files by hand after committing. Any change invalidates the SHA-256 signature and will block deployment. If you need to adjust the logic, submit a new forge request and re-verify.
Commit your strategy files before proceeding to backtesting: 
git add server/strategies/full/SmaCrossover/
git commit -S -m "feat(strategy): add SmaCrossover v1.0.0"