One workflow
Define configurations, validate assumptions, source market data, simulate, optimize, compare, reproduce, generate signals and operate trading workflows.
Research Engine
One environment.Every experiment.
The QuantForge Research Engine holds the study together: market data, simulation, optimization, machine learning, trade-level analytics, and the path to signals. An embedded agent makes that depth usable.
One workflow
Define, validate, run, optimize, analyze, and operate without leaving the environment.
Agent inside
Ask, inspect, adjust configurations, and run supported research from the page you are on.
Engineers on call
Quant engineers stay available when the work needs human judgment.
The proposition
From signal to execution.
A research platform is more than a backtesting library. It is the infrastructure around every experiment, decision and deployment.
Deep observability
Inspect study and trial outcomes, risk-adjusted performance, individual orders, trades, debug data, code snapshots and experiment-to-experiment differences.
AI at your side
A context-aware Codex agent helps explain the interface, navigate the system, inspect research, adjust configurations and run complex authorized workflows.
The practical difference
Instead of stitching together scripts, notebooks, services and dashboards, researchers work within a single repeatable environment. The agent makes that depth more approachable without removing the rigor.
01 / Research lifecycle
Repeatable research, end to end.
Named configurations and reproducible artifacts help move an idea through controlled experimentation toward operating signals.
Define
Strategies, universes, models and experiment intent
A completed experiment retains its resolved configuration, universe membership, relevant resources, logs, study information and strategy-package source snapshot.
02 / Agent integration
An expert companion, inside QRE.
A deep research platform can take years to master. The integrated AI agent makes the platform easier to learn and substantially faster to operate.
Page-aware by design
The agent receives the current page context, so questions and assistance can be grounded in the workspace, configuration or analysis the researcher is viewing.
Ask, understand, act
Ask what settings mean, why a trial behaved a certain way, which research steps to take next, or how to configure and validate a new experiment.
Connected to the platform
Codex app-server integration and QRE-specific MCP tools expose research operations and information to the agent. Domain skills guide more specialized workflows.
More than a chatbot
The agent can inspect results, work with code and configurations, carry out multi-step research workflows and help investigate anomalies and failures.
02 / Agent integration
From question to validated workflow.
Natural-language assistance is most valuable when it is connected to the same mechanisms that researchers use manually.
AI does not replace deterministic calculations or pre-flight validation. Research actions remain constrained by platform capabilities, user permissions and appropriate human oversight.
02 / Agent integration
Compare experiments in conversation.
Select experiments or individual trials and use the agent to investigate differences, hypotheses, trade-offs and sources of performance variation.
Illustrative agent workflow
Agent response
Trial 42 keeps the same entry logic as trial 18. The lookback tightened from 48 to 21 bars, and the stop widened from 1.4 to 1.9 ATR. The objective moved from 0.61 to 0.74, and realized win rate from 58% to 67%. Fees are about 18% higher, and the worst trade deepened from -1.4% to -1.9%. The improvement shows up in the trending half of the sample, not in the chop.
03 / Analytics
Every level of a study, explorable.
Analytics are not a final report. They are an interactive investigation environment spanning experiments, optimization trials and portfolio mechanics.
Drill down, then ask why
From the experiment leaderboard to a specific trial, from equity curve to a single losing order, the system keeps the evidence close. The agent can work with the selected experiment or trial and help frame the comparison in plain language.
Trades, returns and portfolio behavior.
Zoom from performance summaries into orders, executed trades and the market context surrounding a position.
Custom diagnostics and strategy context.
Inspect equity curves and custom model or strategy traces in the same analytical workspace, including per-asset regime metrics and derived signals.
Order-level traceability.
Go beyond aggregate performance to examine when orders were signaled, created and filled, and what happened at exit.
Trial 202 · order details
12 orders · 6 closed positions
Net PnL
+678.37
| Fill | Side | Dir | Stop | Size | Price | Fees | PnL | Return |
|---|---|---|---|---|---|---|---|---|
| 10 Jan 2025 20:00 | Sell | Short | — | 144.01 | 694.25 | 20.00 | +455.72 | +0.46% |
| 12 Jan 2025 12:00 | Buy | Short | TSL | 144.01 | 690.81 | 19.90 | +455.72 | +0.46% |
| 23 Jan 2025 05:00 | Buy | Long | — | 146.37 | 686.17 | 20.09 | -873.84 | -0.87% |
| 24 Jan 2025 23:00 | Sell | Long | TSL | 146.37 | 680.47 | 19.92 | -873.84 | -0.87% |
| 26 Jan 2025 08:00 | Sell | Short | — | 144.33 | 689.84 | 19.91 | +513.40 | +0.52% |
| 26 Jan 2025 10:00 | Buy | Short | TSL | 144.33 | 686.01 | 19.80 | +513.40 | +0.52% |
| 27 Jan 2025 21:00 | Buy | Long | — | 148.72 | 672.91 | 20.02 | +1,276.56 | +1.28% |
| 28 Jan 2025 05:00 | Sell | Long | TSL | 148.72 | 681.76 | 20.28 | +1,276.56 | +1.28% |
| 28 Jan 2025 08:00 | Buy | Long | — | 149.51 | 677.90 | 20.27 | -1,709.84 | -1.69% |
| 28 Jan 2025 19:00 | Sell | Long | SL | 149.51 | 666.73 | 19.94 | -1,709.84 | -1.69% |
| 29 Jan 2025 02:00 | Buy | Long | — | 148.89 | 669.24 | 19.93 | +1,016.37 | +1.02% |
| 29 Jan 2025 05:00 | Sell | Long | TSL | 148.89 | 676.33 | 20.14 | +1,016.37 | +1.02% |
Order 0 · position 0 · Closed
Entry trade 0
Order metadata
Side, size, price, fees, stop type and execution indices.
Trade linkage
Entry and exit trade IDs, direction, status and PnL.
Metrics that reveal the trade-offs.
Performance, risk, activity, exposure and benchmark comparisons are inspectable for individual backtest and trial results.
Result
1 Jan 2025 01:00 – 31 May 2026 23:00
513d 12h
Total Return
+12.89%
Sharpe Ratio
1.51
Max Drawdown
2.71%
Realized Win Rate
73.33%
Window
| Start Index | 1 Jan 2025 01:00 |
|---|---|
| End Index | 31 May 2026 23:00 |
| Total Duration | 513d 12h |
Account value
| Start Value | 100,000.00 |
|---|---|
| End Value | 112,893.41 |
| Min Value | 99,006.86 |
| Max Value | 114,456.34 |
| Invested Capital | 3,195,075.07 |
Returns
| Total Return [%] | +12.89% |
|---|---|
| Benchmark Return [%] | +0.44% |
| CAGR [%] | +8.98% |
| Invested Return [%] | +0.40% |
Risk
| Max Drawdown [%] | 2.71% |
|---|---|
| Max Drawdown Duration | 126d 2h |
| Max Gross Exposure [%] | 102.64% |
| Position Coverage [%] | 2.66% |
| Sharpe Ratio | 1.51 |
| Sortino Ratio | 2.48 |
| Calmar Ratio | 3.32 |
| Omega Ratio | 1.31 |
Trading
| Total Trades | 30 |
|---|---|
| Total Orders | 60 |
| Total Fees Paid | 1,274.28 |
| Realized Win Rate [%] | 73.33% |
| Total Win Rate | 73.33% |
| Best Trade [%] | +2.18% |
| Worst Trade [%] | -1.69% |
| Avg Winning Trade [%] | +1.00% |
| Avg Losing Trade [%] | -1.20% |
| Avg Winning Trade Duration | 11h 10m |
| Avg Losing Trade Duration | 10h 15m |
| Profit Factor | 2.26 |
| Expectancy | 429.78 |
Understand the search, not just the winner.
Optimization diagnostics help researchers see parameter interactions, promising regions and sensitivity of objective values.
03 / Analytics
Designed for heavy research workloads.
The depth of the analytics is matched by the infrastructure required to keep it usable across large histories and many trials.
High-density charts
Dynamic resampling keeps long time series navigable. Date-window plot splitting enables detailed inspection of selected periods, synchronized with debug figures.
Structured debug artifacts
Implementations can emit named debug datasets and Plotly figures. Tabular data is stored efficiently in Parquet and loaded only when requested.
Reproducible comparisons
Compare historical and current configurations; inspect strategy-package changes; identify parameter, universe, data or source-code drift.
Robustness and generalization
Review chronological cross-validation splits, per-split portfolios, time-window stability and objectives that penalize inconsistent performance.
04 / Optimization
A research engine, not a parameter loop.
Optuna studies are persistent, configurable and connected to the surrounding data, validation and analysis infrastructure.
Search modes
Multivariate TPE, random and grid search; numeric, categorical and nested configuration ranges.
Alpha / beta studies
Outer structural choices create inner optimization studies; assess experiment designs alongside their optimized outcomes.
Durable studies
Continue interrupted studies, soft-stop research, checkpoint artifacts and manage storage growth.
Financial objectives
CAGR, returns, Sharpe, Sortino, Calmar, drawdown, win rate, order count and other supported measures; multi-objective selection.
Cross-validation
Chronological splits, warm-up-aware scoring, split-level metrics and robustness-aware objective functions.
05 / Machine learning
ML is part of the strategy lifecycle.
Train and optimize models within the same ecosystem used to test strategies and assess real portfolio outcomes.
06 / Data + universes
Research starts with dependable inputs.
Data acquisition, caching and dynamic universe membership are first-class elements of the experiment definition.
Multi-source data
Binance, Bybit, Alpaca, Tiingo, Yahoo Finance, Databento and Crypto Lake, including selected hybrid live paths.
Gap-aware cache
Serve covered ranges locally, fetch only missing intervals, append new bars and reuse finer-resolution cached data.
Composed datasets
Multiple timeframes, warm-up histories, raw and Heikin-Ashi feeds, optional sub-datasets and alternate market features.
Programmable universes
Static or CSV membership, ranked dynamic selection, symbol translation, date bounds and extensible eligibility rules.
07 / Operations
Designed to run real experiments.
Queues, service monitoring and reproducibility help researchers move beyond fragile single-session workflows.
Queue and observe
Redis/RQ priority queues, job status, captured logs, stop controls, ML fan-out and failure diagnostics.
Retain the evidence
Persist configs, universe selections, code snapshots, study history, metrics and selected artifacts.
Signals and trading
Shared research foundations for signal generation, persisted signal state, market actions, orders, fills and operational controls.
Control cost and memory
Market-data reuse, compiled vectorbtpro and Numba pathways, optional debug collection and retention policies.
08 / What this changes
Depth without unnecessary friction.
The platform remains technically deep. The agent changes how quickly that depth can be discovered, understood and put to work.
For the researcher
Explore rigorous experiment analysis, ask questions in context, compare selected trials, adjust configuration ranges, validate proposed changes and get help driving multi-stage research workflows.
Research
Test hypotheses, examine robustness and preserve reproducibility.
Build
Connect machine learning, configuration and strategy workflows.
Operate
Track studies, signals, queues and system health.
QuantForge Software
One engine.An expanded research team.
From a single parameter change to an agent-assisted investigation across studies, models, trades and code: the same research infrastructure supports the entire journey.
Implementation note: capabilities described reflect the supplied 2026 guide and stated agent integrations. Integration availability depends on provider credentials and deployment. Trading actions require client-specific verification, risk controls and operational approval.
At your service
You do not need to be a quant engineer.
The platform is deep, but doesn't need to be overwhelming. Bring the strategy, even if it's rough. The engineering is our job.
Quant engineers on call
We are quant engineers at your service. Implementation, research, optimization, and live operation stay available as human work whenever you want that judgment in the loop.
The agent covers the hours
Human time is the expensive part. The embedded AI agent explains the system, inspects results, adjusts configurations, and runs supported workflows, so engineer hours are reserved for the decisions that need them.
Use us when you want a quant. Use the agent when you want the work done without starting a timesheet.
Pricing and next steps
Start once. Then pay for the machine you actually need.
Before you commit
Free 30-minute consultation
Thirty minutes to see if we're the right fit. Tell us what you're trying to build, and we'll tell you whether this is the right next step.
To get started
01Startup fee
$250
once
The consultation, the deployment, and a 30-minute system walkthrough are one fee.
Initial consultation to scope the setup
Deployment of your environment
30-minute system walkthrough
Consultation, deployment, and a 30-minute system walkthrough are included.
Ongoing access
02Monthly fee
$100–$300
per month, typical
The fee follows the virtual machine and the data sources your use case needs.
VM size and data-source needs set the actual number
Most setups land between $100 and $300 a month
Quoted for your workload, not from a fixed tier menu
We size this from the VM and data sources your use case needs.
Beyond the included walkthrough
03Human time
$75
per hour
Quantitative work after the initial consultation and walkthrough.
The consultation and walkthrough are already in the startup fee
Further research, implementation, and support are billed by the hour
The embedded agent covers day-to-day platform work without a timesheet
Billed only for work beyond the included consultation and walkthrough.