Join the Studio waitlist

Backtest locally today — Docker only, no key, free for personal trading. Drop your email to get early access to hosted Studio and Optuna optimization. One launch email and the occasional progress note.

See it match TradingView trade-for-trade ↗PineForge on GitHub

Hosted Studio launches Q4 2026. The local runtime is free for personal trading today.

Try it · 2 min

no key · runs local

01Docker is all you need

No install, no key, no signup. The PineForge runtime is one container image — transpile and backtest both run inside it, offline.

02Transpile your Pine to C++

bash

docker run --rm --network=none \
  -e PINEFORGE_TRANSPILE_ONLY=1 \
  -v "$(pwd)/strategy.pine":/in/strategy.pine:ro \
  ghcr.io/pineforge-4pass/pineforge-engine:latest > strategy.cpp

03Backtest against your OHLCV CSV

bash

docker run --rm --network=none \
  -v "$(pwd)/strategy.cpp":/in/strategy.cpp:ro \
  -v "$(pwd)/ohlcv.csv":/in/ohlcv.csv:ro \
  ghcr.io/pineforge-4pass/pineforge-engine:latest > report.json

View the codegen on GitHub Browse 246 backtests we ran (gallery)

245/246

strict TV parity

trade-for-trade · 246 reference strategies · 0 engine bugs

100/100

vs PyneCore 85/100

three-engine bake-off; PineForge leads

~98%

strategy coverage

of what real Pine v6 scripts use; full table on /coverage

375k+

trades validated

aggregate across the 246-strategy corpus

Open core

Apache engine · runs on your machine

audit the engine · run locally · free for personal trading

strategy.pine→libgenerated.so→report.json

compile cache: HIT·4.2k bars / 38ms

PineScript v6

//@version=6
strategy("EMA Cross", overlay=true, initial_capital=10000)

length = input.int(14, "Length")
sig    = ta.ema(close, length)

if ta.crossover(sig, sig[1])
    strategy.entry("long", strategy.long)

if ta.crossunder(sig, sig[1])
    strategy.close("long")

Generated C++ · emitted by codegen

class GeneratedStrategy : public BacktestEngine {
    ta::EMA          _ta_ema_1{14};
    Series<double>  _s_sig{500};

    void on_bar(const Bar& bar) override {
        int    length = get_input_int("Length", 14);
        double sig    = _ta_ema_1.compute(bar.close);
        _s_sig.push(sig);

        if (sig > _s_sig[1] && _s_sig[1] <= _s_sig[2])
            strategy_entry("long", true);
        if (sig < _s_sig[1] && _s_sig[1] >= _s_sig[2])
            strategy_close("long");
    }
};

Trades

142

Net P&L

+$3,184.62

Sharpe

1.48

Max DD

−12.7%

Validated against TradingView exports

ma-cross paritysupertrend parityinside-bar paritystochastic-slow paritypivot-ext parity4ema-rsi parityliquidity-sweep paritymarket-shift paritymacd-histogram paritydonchian-breakout paritymatrix-pca parityregex-filter parity

Why PineForge

Three things TradingView’s runtime won’t give you.

01 / DETERMINISM

Same script, same data, same trades. Byte-reproducible.

02 / SPEED

Native C++. 50k bars in tens of milliseconds.

03 / OWNERSHIP

Your data, your machine, your strategy. Offline by default.

For AI assistants

Use PineForge from Claude, Cursor, or any MCP client.

Our Model Context Protocol server runs as one container — your AI agent transpiles Pine to C++ and backtests on your data, all inside it. Talk to your strategy in natural language. No API key; your code and data never leave your machine.

Run (one command)OHLCV stays on your machine.

docker run --rm -i -v "$PWD:/work" ghcr.io/pineforge-4pass/pineforge-codegen-mcp:latest

Tools exposed

·transpile_pine — Pine v6 → C++
·backtest_pine — backtest on your CSV
·backtest_pine_grid — sweep params in one shot
·fetch_binance_ohlcv — pull market data

See full setup →

Where PineForge sits

One runtime that occupies a corner no one else does.

Five axes that matter to a quant who wants to ship a strategy as a product. PineForge is built around all five.

Axes chosen to test PineForge's thesis. Scored from public docs and benchmarks. Methodology →

PineForge vs TradingView

WINS

5 / 5

SPEED

PRIVACY

LICENSE

OSS

DATA

PineForge vs MQL5 Market

WINS

4 / 5

SPEED

PRIVACY

-1

LICENSE

OSS

DATA

PineForge vs QuantConnect

WINS

4 / 5

SPEED

PRIVACY

LICENSE

OSS

DATA

1 tie

PineForge vs Backtrader

WINS

3 / 5

SPEED

PRIVACY

LICENSE

OSS

-2

DATA

1 tie

Differentiation scores by engine and axis
Engine	Native speed	Source privacy	License control	OSS audit	Data freedom
PineForge	5/5	4/5	5/5	3/5	5/5
TradingView	2/5	2/5	1/5	0/5	2/5
MQL5 Market	4/5	5/5	3/5	0/5	3/5
QuantConnect	3/5	1/5	1/5	3/5	4/5
Backtrader	1/5	0/5	0/5	5/5	5/5

Capabilities

Five things you can’t do on TradingView alone.

shipped

Bring your own data

Run any custom data — daily, intraday, alt — without TradingView's symbol restrictions.

shipped

Intra-bar resolution at any cadence

Walk inside the bar with sub-bar resolution. No more last-tick approximation.

building Q3 2026

Optuna with a custom objective

Optimize with any objective function — Sharpe, drawdown, profit factor, your own.

building Q3 2026

Forward-test with TradingView-shaped webhooks

Walk-forward analysis built in. Out-of-sample is the default, not an afterthought.

designed 2027

Strategy marketplace · sell compiled .so

Sell compiled .so. Time-bound, machine-bound, broker-bound — set by you, not by a platform.

Bring your own dataexpand

Any OHLCV CSV — your tick data, your custom feed, your alternative-asset history. Run offline, in CI, in Docker. No upload, no API key.

Intra-bar resolution at any cadenceexpand

Six distribution modes (uniform, cosine, triangle, endpoints, front/back-loaded), optionally volume-weighted. A limit at $100 inside a 95–105 bar fills at exactly $100 — TradingView’s bar magnifier without the subscription.

Optuna with a custom objectiveexpand

Sharpe, Sortino, drawdown, profit factor — or any one-line lambda you want the optimizer to chase.

def objective(report):
  return 0.6 * report.sharpe - 0.3 * report.max_dd + 0.1 * report.profit_factor

Forward-test with TradingView-shaped webhooksexpand

Drop-in replacement for TradingView alerts. Same JSON shape, same runtime as your backtest — no rate limits, no replay drift.

Strategy marketplace · sell compiled .soexpand

AES-256-GCM-encrypted shared library. Ed25519-signed, machine-bound, time-bounded licenses. Buyers tune the inputs you expose; they never see the source.

How it works

Four stages.
Zero hidden state.

01 / input

.pine source

your file

02 / parse

AST + type-check

Pine v6 spec

03 / transpile

C++ codegen

no runtime tax

04 / compile + run

Native binary

~9.7ms / 50k bars

05 / validate

Trade-for-trade

245/246 strict · 0 engine bugs

01 · input

.pine source

your file

02 · parse

AST + type-check

Pine v6 spec

03 · transpile

C++ codegen

no runtime tax

04 · compile + run

Native binary

~9.7ms / 50k bars

05 · validate

Trade-for-trade

245/246 strict · 0 engine bugs

Studio preview

One report, every number you actually trust.

ETHUSDT · 15m2025-07 → 2026-04EMA Cross · L=21

Net

+31.84%

Sharpe

1.48

Max DD

−12.7%

Win rate

52.1%

Trades

142

Profit factor

1.71

Equity curve · vs Buy & HoldPineForgeB&H

Jul '25Sep '25Nov '25Jan '26Mar '26Today

PineScript v6 coverage

~90% of the language.
~98% of what strategies actually use.

Drawing primitives, alerts, and live-tick semantics are out of scope by design — PineForge runs offline. Everything that determines a trade is in.

Full function-by-function coverage

ta.* — 59 functions + 8 series + pivot helper (67 classes)	68/68	100%
math.* — deterministic + rolling	core	supported
str.* — format · split · regex · tostring	core	supported
strategy.* — orders · accessors · risk gates	full	100%
array⟨T⟩ · map⟨K,V⟩ · UDT	via codegen	supported
matrix⟨T⟩ — Eigen-backed	50+ ops	supported
request.security — ratio · calendar · lower-TF	core	supported
drawing & alerts	—	out of scope

TradingView parity

We don’t ask you to trust the engine.
We diff it.

Every release is validated trade-by-trade against TradingView’s CSV exports — 246 reference strategies, full corpus open-source. 245 strict, 1 documented TV-side anomaly, zero engine bugs.

246/246

all accounted for — zero engine bugs

Excellent245TV-side anomaly1Engine bugs0

Three-engine bake-off · 100 strategies · same OHLCV

How PineForge stacks up against PyneCore and PineTS.

Full comparison & reproduction recipe

C++ static lib

PineForge

100 / 100

100 excellent · 0 strong · 0 weak

Python (cloud-transpiled)

PyneCore

85 / 100

85 excellent · 2 strong · 10 moderate · 3 weak

TypeScript (LuxAlgo)

PineTS

indicators only

strategy backtester is upstream roadmap

The 15-strategy PyneCore gap is consistently in bracket / trail / partial-exit semantics, where PyneCore’s broker emulator differs from TV. Three canonical examples: 06-liquidity-sweep, 07-scalping-strategy, 49-partial-exit-qty-percent. See the per-strategy breakdown →

What’s coming

Backtest is the start. Live is the point.

now
shipped
Deterministic batch backtests
Pine v6 → C++ → trades. 245/246 strict TV parity, 0 engine bugs, 93.06% line coverage, open-source pineforge-engine.
Q3 2026
building
Forward-test + Optuna optimization
Streaming feed_bar() API. Multi-window robustness scoring.
Q4 2026
next
Hosted Studio · paper trading
Code · Backtest · Optimize · Compare workspace. One broker bridge at launch.
2027
later
Live execution · marketplace
Multi-broker fills. Encrypted strategy distribution. Audit log per order.

Built for solo quants first.

Solo quant traders

Three things TradingView’s runtime won’t give you.

Same script, same data, same trades. Byte-reproducible.

Native C++. 50k bars in tens of milliseconds.

Your data, your machine, your strategy. Offline by default.

Use PineForge from Claude, Cursor, or any MCP client.

One runtime that occupies a corner no one else does.

Five things you can’t do on TradingView alone.

Four stages.
Zero hidden state.

One report, every number you actually trust.

~90% of the language.
~98% of what strategies actually use.

We don’t ask you to trust the engine.
We diff it.

Backtest is the start. Live is the point.

Be there when Studio opens.

PineScript backtests,deterministic,on your data.

Built for solo quants first.

Solo quant traders

Three things TradingView’s runtime won’t give you.

Same script, same data, same trades. Byte-reproducible.

Native C++. 50k bars in tens of milliseconds.

Your data, your machine, your strategy. Offline by default.

Use PineForge from Claude, Cursor, or any MCP client.

One runtime that occupies a corner no one else does.

Five things you can’t do on TradingView alone.

Four stages.Zero hidden state.

One report, every number you actually trust.

~90% of the language.~98% of what strategies actually use.

We don’t ask you to trust the engine.We diff it.

Backtest is the start. Live is the point.

Be there when Studio opens.

Four stages.
Zero hidden state.

~90% of the language.
~98% of what strategies actually use.

We don’t ask you to trust the engine.
We diff it.