Finforge is an AI strategy platform powered by sophon‑n3.0, an always-learning model. Start with clean, point-in-time market data and chain nodes like the Screener and Forecast nodes (using sophon‑n3.0) into flexible pipelines to output a tradable portfolio. Connect your broker to go from paper to live trading.
See how each project is built and how it performs. Every preview includes the sophon-n3.0 model node tree, a growth chart for benchmark index comparison and a table with YoY metrics. There are more projects on our Live demos page.
Design autonomous strategies by connecting purpose-built nodes. Each node performs a specific task, like screening assets, forecasting prices, constructing portfolios, or visualizing results. Allowing you to create sophisticated, AI-driven trading systems without writing code.
node-1
Assets
You start every recipe by choosing your desired assets: Stocks, crypto, FX or indices. Supports intraday 1-hour bars and end-of-day data, with survivorship-bias-free equity universes (incl. delisted constituents) for accurate backtests.
node-2
Screener
(using sophon-n3.0)
The sophon-n3.0 Screener node filters the asset universe with learned thresholds (e.g., Revenue YoY > x%) tuned to your objective, like maximizing information ratio vs. a benchmark. It recalibrates each timestep via continual learning.
node-3
Forecast
(using sophon-n3.0)
The sophon-n3.0 Forecast node predicts asset properties (e.g., price or news sentiment score) at fixed horizons (days or weeks) and adapts each timestep via continual deep learning, using ensembles and model chains to stay accurate as markets shift.
node-4
Make portfolio & Backtesting
This node turns upstream signals into a tradable portfolio (target weights & orders). Choose long/short, margin, constraints, and rebalancing. Backtests model broker fees, margin calls and account for short borrow costs.
node-5
Trade live
Consumes target weights from Make Portfolio and integrates with supported broker trading APIs (Alpaca Markets, Interactive Brokers and Binance support under development). Tracks orders and positions in real time and supports paper trading where available.
node-6
Dashboard
Visualize outputs from any node, like Asset, Screener, Forecast, Make Portfolio, or Trade Live. Glanceable, real-time tracking of signals, forecasts, weights, orders, positions, and drawdown. Visualizes data fetched from supported brokers.
Traditional backtests often overfit and static models underfit when regimes change (“concept drift”). sophon-n3.0 counters both with continual/online learning and an in-node mixture-of-experts: parallel/stacked sub-models plus a gating network that learns which expert to trust when (per asset, per timestep), updating online as data shifts. This lets each sophon-n3.0 Screener node, and sophon-n3.0 Forecast node adapt quickly without catastrophic forgetting while preserving long-range structure.
Each sophon-n3.0 node blends parallel deep models (an ensemble) and stacked sub-models in sequence. The ensemble reduces variance and improves uncertainty calibration. The in-node chain lets later sub-models learn from earlier outputs for regime-aware, multi-step decisions. A learned gate picks which expert to trust at each timestep.
Markets drift. sophon-n3.0 updates each timestep to track new patterns while preserving prior knowledge, allowing it to address the stability–plasticity trade-off that breaks static and one-shot models.
sophon-n3.0 can effectively integrate many data streams like prices, news and exogenous data like oil & gold futures as market sentiment, to make a forecast for any asset
We source market data from exchange-connected providers and deliver survivorship-bias-free universes (incl. delisted names). Corporate actions (splits/dividends) are aligned into intraday/EOD bars and fundamentals/news are stored point-in-time to prevent look-ahead. We also run automated outlier/anomaly checks and missing-bar repair using exchange calendars, and standardize/scale features so they’re ML-ready for continual learning.
Sourcing live financial market data for AI models is hard. Finforge uses providers with servers co-located at exchanges, aligns all data in time correctly and readies it for machine learning.
Can I sign up for Finforge right now?
Finforge is currently in closed beta. We expect to open for sign-ups in the coming months. If you're interested in applying for early access, you can use the link in the top right corner and fill out our application form.
Which brokerages does Finforge support?
Finforge supports Alpaca.markets and is currently working on implementing support for Interactive Brokers and Binance, with more additional broker support later in our roadmap.
How much does Finforge cost?
During early access, Finforge is free for a limited number of users.
When we launch publicly, it will be free to register and view projects (“recipes”).
Building your own recipes will require a paid subscription and may include usage-based compute costs.
Who is Finforge for?
Finforge is designed for both technical users and ambitious retail traders. Including people coming from platforms like TradingView or Composer. If you’re interested in how you can use AI to build robust and efficient trading strategies, we think you're going to find Finforge interesting.
Is there a free trial available for Finforge?
If you want to try Finforge early, we recommend applying for early access.
After launch, we plan to offer a free tier with limited functionality.
How does Finforge compare to other tools?
Most tools fall into one of two categories:
Finforge is distinctly different from both.
It is powered by sophon-n3.0, our proprietary financial AI model trained specifically on market and trading data. Our node-based editor allows users to build far more sophisticated and reliable strategies than manual rule-based tools, or LLM prompting.
Do I need to know how to code to use Finforge?
No. Finforge is fully no-code, built around a visual node-based editor.
You can build complete trading strategies without writing a single line of code.
What types of assets can I trade using Finforge?
Finforge supports building recipes using five asset classes:
Stocks, Crypto, Forex, Indices, and Indicators.
Live-trading availability depends on your connected broker.
How does Finforge approach security?
Finforge is built with industry-strength security by design: all data is encrypted in transit and at rest, and access is tightly controlled on a least-privilege basis. Your data and projects are fully isolated and are never shared with anyone unless you explicitly choose to do so.
