Fintech · Forge Strategy Lab, a Prolink venture

A trader designs, backtests, paper-tests and goes live on their own broker without writing code, and the platform never holds their money.

Retail and prop-firm gold traders want to prove a strategy works before risking money, but building, testing and running one usually means writing code and wiring up a broker yourself.

Talk to us about something similar All case studies

ClientForge Strategy Lab, a Prolink venture

SectorFintech

WhenAlpha, 2026

Used byMembers, support admins, admins

Plugs intoOANDA, MetaTrader 5 (own connector), Stripe, Postmark, Anthropic Claude, Forex Factory

AIA page-aware assistant that answers from a live catalogue of what the engine can do, a four-agent strategy optimiser with a risk officer holding a hard veto, an AI support agent, and a morning review email. Three models routed by task, every call metered against a per-user budget.

What we built

Wizards, a builder, a backtester, a paper trader, live connectors and an AI layer with a budget.

Sixteen guided strategy wizards plus a free-form builder, both compiling to one format so a single testing engine serves both.

Backtesting against minute-by-minute history with profit, win rate, drawdown, a day-by-day breakdown and a full trade log, with large runs queued to a background worker.

Paper forward testing against the live market on a fifteen-second tick, pausable and resumable.

Go-live through a connector installed beside the member's own terminal or through the broker's API, with each live strategy pinned to exactly one connector.

Prop-firm guard rails that stagger timing, skip the odd entry and jitter size when members run near-identical strategies at the same firm.

Ops safety nets: a critical-pages watch that phone-pushes on a burst of errors, connector health checks, and a queued email worker with retries and bounce suppression.

ai safety

The anti-hallucination contract

The assistant's knowledge of the product is regenerated on every request from the same catalogue that drives the product, so it cannot describe a feature that does not exist.

cost

Hard budget rails

Per-user monthly caps by tier, purchasable top-ups, a per-run ceiling that throws when exceeded, and a cheap classifier deciding when the expensive model is justified.

fairness

Prop Shield

Staggers, skips and jitters trades scaled by how many clones exist at one firm, leaves a lone account untouched, and discloses itself per trade.

How it plugs in

Instructions go to the member's own account. Never money.

The platform sends signals to the member's terminal or broker. It holds no funds and takes no custody, by design.

The applicationOANDAMetaTrader 5 (own connector)StripePostmarkAnthropic Claude

job Forward tick about every fifteen seconds walking every active session, with a broker executor every minute and hourly reconciliation

job A regime snapshot every minute classifying the market, and news protection pausing sessions around high-impact events

job Flood-safe alerting with high-water marks and hard caps, born from a real incident of hundreds of emails about one resolved problem

What changed

A non-technical trader can design, backtest, paper-test and go live on their own broker account without code. The platform never holds customer funds.

one detail only this project hasThe deployer refuses to ship unless the commit is pushed and a changelog entry exists, and runs drift checks on typography, engine schema and migration numbers before upload.

Something similar?

Building AI features that need a budget and a leash?

Book a walkthrough Request a callback