P2 · Strategy coder
Describe the strategy. Read the code that trades your money.
the AI writes code you review — never a signal you obey. import a Pine script, paste a PDF of your notes, describe it in plain English. what comes out is Python you can read, edit, backtest and version like software. that is the white-box positioning, and it is the only reason a compliance-forward retail algo product can exist under the SEBI framework.
in the terminal
plain english → spec → python · same IR as the templates librarybuy nifty on the 15-minute opening-range breakout when the breakout candle has above-average volume. exit on a 1% trailing stop, or at 3:00pm ist. paper only, one lot.
from vq import Strategy, indicators as ind
class OpeningRangeBreakout(Strategy):
instrument = "NIFTY-FUT"
timeframe = "15m"
session = "09:15-15:00"
def setup(self):
self.or_high, self.or_low = None, None
self.avg_vol = ind.rolling_mean("volume", window=20)
def on_bar(self, bar):
if bar.time.time() == time(9, 30): # first 15m closed
self.or_high, self.or_low = bar.high, bar.low
return
if self.position.is_flat and self.or_high is not None:
if bar.close > self.or_high and bar.volume > self.avg_vol.value:
self.buy(1, stop=trail(pct=0.01), reason="orb+vol")
if bar.time.time() >= time(15, 00) and self.position.is_long:
self.close(reason="eod-exit")what ships
- plain english → spec → codemvp
- the pipeline is three steps, not one: extract a structured spec, then compile to Python. you see both. when the model is wrong, it's wrong in a place you can point at.
- PDF and screenshot ingestionmvp
- paste a strategy write-up, a broker research note, a screenshot of an idea from twitter. we extract the spec and generate the code.
- importer — Pine, AFL, Pythonmvp
- every serious trader already has strategies somewhere. paste Pine Script, Amibroker AFL or an existing Python file and get a working strategy on our IR — with an honest fidelity report showing what converted cleanly, what was approximated and what needs manual review.
- code editormvp
- Monaco-based, schema-aware autocomplete on the strategy API surface. types check as you type; the compiled strategy has to typecheck before it can backtest.
- template librarymvp
- the ~30 patterns most retail strategies are variants of — MA crossover, breakout with volume, RSI mean reversion, supertrend, opening-range breakout, time-triggered straddle. every template is code, published, never black box.
- version controlmvp
- git-style history per strategy. diff any two versions, roll back to any tag, name a version at the moment you deploy it.
- visual block builderv1
- drag-drop authoring that compiles to the same IR as everything else. for people who prefer to see the graph.
- parameter sweepsv1
- run across a parameter grid; overfitting warnings up-front for the whole sweep, not just the best result.
- options payoff builderv2
- multi-leg strategies with a live payoff visualizer.
non-negotiables
generated code, sandboxed
no network, no filesystem, no arbitrary imports. the runtime is what the platform provides; nothing else runs.
no auto-deploy
a strategy that has not backtested with a human diff review cannot deploy live. LLM output is never the last line of defence.
cache by intent
two users asking for the same thing in different words hit one generation. cost per strategy drops by an order of magnitude, and the code is the same reviewed template every time.
the other four surfaces
waitlist
start when we're ready.
we'll email once when there's something to sign up for. no drip, no digest.