VegaTrade

Every AI decision, replayable.

The audit-chain-native AI HFT platform for hedge funds, institutional trading desks, and family offices.

Glass-box AI decisioning in the slow loop. Deterministic C++ in the hot path. Every signal, every order, and every strategy shows its work — and can prove it years later.

Request Access

Differentiation

An AI HFT platform built audit-chain-first. Glass-box AI. Deterministic execution.

Glass-Box AI

Every decision shows its work

The AI is a glass box, not a black box. Every AI decision — strategy selection, parameter choice, signal generation, lot disposition — captures its prompt, model identifier, temperature, inputs hash, output, and reasoning into the audit chain at the moment of decision. The Replay API reconstructs any past decision byte-identical years later. Black-box AI is forbidden by architecture, not by policy.

WORM. Hash-Chained. Signed.

17a-4-aligned audit architecture

Every event is appended to a per-tenant regulatory chain with SHA-256 row hashing and daily-root ECDSA-P256 signing in a hardware security module. Chain integrity is enforced in code, verified at insert time, and replayable byte-identical from cold immutable storage. The audit log is a primary data structure; losing events is treated as a critical bug, never a performance trade-off.

AI Proposes. Humans Approve.

Supervised AI strategy lifecycle

AI authors bounded strategy parameterizations — never opaque code — and every proposal lands in an approval queue where a human operator reviews, approves, or rejects it. The approval, the approver, and the reasoning are all audit-chain events. Per-symbol trade decisions run inside those approved bounds with per-decision reasoning traces. The AI never grants itself authority.

§1091 Wash-Sale. Form 8949.

Tax-lot tracking + harvesting backbone

Lot accounting (FIFO / LIFO / HIFO / Specific-ID), §1091 wash-sale detection with basis adjustment and holding-period tack, Form 8949 row generation, and 1099-B reconciliation are landed. Per-disposition Specific-ID override is audit-chain emitted with operator attribution. User-facing tax-loss-harvesting workflows are the next layer.

Backbone built — user-facing TLH workflows on the roadmap

Discuss timing →

Glass-Box Strategy

Validity-invariant gates · open methodology

Every strategy is a glass box: open code, named parameters, reproducible backtests. Each one clears walk-forward and nested cross-validation, deflated Sharpe, and probability-of-backtest-overfitting gates before reaching paper trading. Execution-model fidelity bands quantify the cost the backtest charges. The platform publishes net-of-all-fees Sharpe alongside gross, and for tax-aware strategies, after-tax Sharpe at the user's marginal rate — the number a fiduciary defends, not the one a marketer picks.

Two Loops · One Boundary

Slow-loop AI · Hot-path HFT execution

The platform separates strategic reasoning from order execution. The slow loop is Python — large-language-model reasoners work on markets, regimes, and signals at seconds-to-minutes cadence. The hot path is C++ — deterministic, pre-allocated, no exceptions, executing in microseconds-to-milliseconds. The two are joined by an ECDSA-P256-signed ZeroMQ boundary. No model call ever touches the order path; every order passes the pre-trade risk gateway.

Who this is for

Built for allocator scrutiny, desk governance, and generational stewardship.

For Hedge Funds

Evidence your next ODD questionnaire asks for

Operational due diligence doesn't take "trust us" for an answer. Every AI decision, every order, every risk-gate verdict, and every administrative action on the platform is a hash-chained audit event with operator attribution — reconstructible byte-identical from immutable storage.

Strategy validation is methodology-first: walk-forward and nested cross-validation, deflated Sharpe, and probability-of-backtest-overfitting gates stand between a backtest and paper trading.

How VegaTrade fits a fund →

For Trading Desks

The forensic surface procurement asks for

Every privileged action — authentication, trade execution, administrative change, tax disposition — is designed to be geo-attributable through the audit chain, with originating IP and derived geolocation captured at the action and joined by chain reference downstream.

A kill switch reaches every fast-loop component in under 100 milliseconds over a signed control channel, with the issuing operator and every in-flight order recorded in the chain. The risk gateway is the sole order path — bypassing it is architecturally impossible.

How VegaTrade fits a desk →

For Family Offices

After-tax discipline as a first-class system

Family capital is taxed capital. Lot-level accounting, §1091 wash-sale detection, Form 8949 row generation, and 1099-B reconciliation are built into the platform's core — not exported to a spreadsheet at year-end.

For tax-aware strategies the platform publishes after-tax Sharpe at the investor's marginal rate alongside the gross figure, so the number under discussion is the number the family actually keeps.

How VegaTrade fits a family office →

Inside the platform

The console an operator actually works in.

These are screenshots of the live paper-trading console — the same code path the production platform runs, filled by a simulated broker. Watch for the part most platforms don't show you: the audit trail and the AI reasoning traces are product surfaces, not internal tools.

VegaTrade AI Decisions table showing per-symbol buy, sell, and hold decisions with the exact model identifier, a reproducibility-class badge, confidence, sizing percentage, and a reasoning-trace button on every row — Every AI decision is a table row with the exact model identifier, a reproducibility class, confidence, sizing — and a reasoning trace one click away. Simulated paper-trading environment with illustrative data. Not real market executions, not real funds, and not a representation of any trading results.

VegaTrade Audit Trail explorer showing the regulatory chain-of-custody event log with chain verification status, event-type filters, producer IDs, and per-event identifiers including AI inference events — The audit-trail explorer: chain-of-custody events — including every AI inference — filterable, exportable, and chain-verified. Simulated paper-trading environment with illustrative data. Not real market executions, not real funds, and not a representation of any trading results.

VegaTrade dashboard with buying power, open positions, P&L, equity curve, and a prominent banner stating the environment is simulated paper trading with no real funds — The operator dashboard: account state, equity curve, risk-limit utilization, and kill-switch state on one surface. Simulated paper-trading environment with illustrative data. Not real market executions, not real funds, and not a representation of any trading results.

VegaTrade Tax Center showing a Schedule D summary with short-term and long-term proceeds, cost basis, section 1091 wash-sale disallowed amounts, and an open tax lots table with per-lot coverage status — The Tax Center: Schedule D rollup, §1091 wash-sale adjustments, and lot-level detail — decimal-exact, in the product. Simulated paper-trading environment with illustrative data. Not real market executions, not real funds, and not a representation of any trading results.

Explore the full platform →

Engineering, By The Numbers

Discipline you can count.

These are engineering facts from the repository as of June 2026 — not performance figures, not traction claims. The platform is pre-launch; the discipline is not.

126
Architecture decision records: 53
Component specifications: 700+
Test files across four languages: 600+
Merged pull requests: 74
Application surfaces in the console: 13
Audit producer domains

Founder & CEO

Jon Toroni

Over 20 years of experience in accounting and finance. Worked with a number of SaaS companies. Software design experience spanning 8 years.

Audit the architecture. Then audit ours.

Request Access