SystemSelf reportedCryptoC++algorithmic-tradingalgotradingbacktestbacktestingc-api

pineforge-engine

Deterministic PineScript v6 backtest runtime. Validated trade-for-trade against TradingView.

System Structure

Data Used
BinanceOHLCV/호가/체결 데이터
Rules / Strategy

저장소별 전략/실행 조건 확인 필요

Execution

백테스트/분석 결과를 바탕으로 수동 판단

Editor Summary

저장소 설명과 공개 메타데이터 기준으로 Python 기반 구현, 거래소 API/프레임워크, 백테스트, TradingView 웹훅 성격의 프로젝트로 파악했습니다. 확인 근거는 README, 저장소 토픽, 저장소 설명, GitHub 지표, 파이프라인 필드이며, 주요 데이터는 Binance, OHLCV/호가/체결 데이터입니다. 전략/실행 조건은 저장소별 문서와 코드 확인이 필요하며, 실행 방식은 "백테스트/분석 결과를 바탕으로 수동 판단"라고 보수적으로 기록했습니다. GitHub 지표는 별 160개, 포크 37개입니다.

Repositorypineforge-4pass/pineforge-engine
Creatorpineforge-4pass
Stars / Forks★ 160 / 37
LicenseApache-2.0
Last Updated2026-07-08
Snapshot Date2026-07-08 (README below is a copy from this date)

This is third-party open-source code. QuantField does not guarantee its behavior or safety. Review the code before installing or running it.

View on GitHub →

README

PineForge

CI Docs License Language Parity Speed codegen on PyPI MCP server

🌐 pineforge.dev · ☁️ Hosted MCP (no install) · 🐳 Run locally (Docker MCP) · 📦 Transpiler (pip) 📖 API Documentation · ⚡ 60-second Tutorial · 🧪 Coverage Map · 🔬 Benchmarks

Backtest PineScript with AI — no build step

Click on below for 60 seconds tutorial

Real backtest for pineforge on Claude in 60 seconds

The fastest way to use PineForge: let your AI agent write, run, and optimize strategies for you via the pineforge-codegen-mcp container — a single self-contained image (built on the pineforge-release image, which bundles the engine runtime and the transpiler) that runs the whole transpile→backtest loop in-process. Pull it, mount a working dir, done. No host build step, no docker-in-docker, no API key — nothing leaves the box.

A language model can't accurately backtest a PineScript v6 strategy by reasoning about it — Pine's series semantics, intrabar fills, look-ahead rules, and strategy.* order logic don't reproduce from approximation, so a hand-simulated trade list or P&L is unreliable and won't match TradingView. This engine runs the real deterministic computation instead: 251/252 reference strategies at strict TradingView parity, 0 engine bugs (methodology & validation, as of 2026-06-29). That is why an agent asked to "backtest this Pine / check if it's profitable / reproduce my TradingView results" should call the MCP tools rather than approximate the run.

The workflow (fully local — source never leaves your machine):

  1. Agent writes (or you paste) PineScript v6 source
  2. The container transpiles Pine → C++ in-process with the bundled [pineforge-codegen](https://g