Architecture

Infrastructure that makes autonomous quant agents credible.

TrustyGenius is not a chat layer bolted onto a broker API. The agents sit on top of a research and execution substrate: market data, graph computation, validation, portfolio state, audit trails, and database-enforced access controls.

Postgres
System of record and policy layer
Rust
Hot-path extensions near the data
IBKR
Paper-first broker integration
VPC
Deployed where your data lives
The loop

From market event to reviewed decision

The platform is built around the full research lifecycle. Each step leaves state behind, so the team can inspect why a strategy was proposed, which evidence supported it, and when it should be paused.

  1. 01

    Ingest

    Market data, broker state, reference data, filings, macro series, and custom feeds land in a shared data plane.

  2. 02

    Compute

    Graph and time-series extensions propagate derived objects, rolling contracts, indicators, risk views, and scenario overlays.

  3. 03

    Research

    Geniuses generate hypotheses, run Python, call approved tools, and write evidence into the journal instead of leaving it in a chat transcript.

  4. 04

    Validate

    Backtests, walk-forward checks, Monte Carlo paths, and paper trading create a record of what survived and what failed.

  5. 05

    Monitor

    Live strategies are watched for regime drift, alpha decay, correlation creep, reconciliation breaks, and risk-limit violations.

Core stack

Postgres-native by design

Trading systems fail when intelligence, data, and permissions are split across too many loosely connected services. TrustyGenius keeps the important state close to Postgres and uses focused extensions where the default database is not enough.

tg_graph

A live DAG for market objects, derived series, strategy dependencies, and price propagation.

tg_timeseries

Columnar time-series storage and Arrow IPC paths for fast research reads.

tg_indicators / tg_pricing

Technical indicators, options pricing, validation helpers, and compute routines near the data.

tg_auth

JWT validation and row-level security at the database boundary, not just in application code.

.NET API

Application services, broker integration, scheduled jobs, SignalR updates, and operational workflows.

React workbench

Research, dashboards, strategy configuration, paper trading, and admin views for the team.

Security and control

Your alpha should not become someone else's training data.

Runs inside your boundary

Deploy in your cloud, your VPC, or on-prem infrastructure. Data sources and broker credentials stay under your control.

Database-enforced isolation

Row-level security and database-side JWT validation reduce reliance on app-layer filtering.

Auditable agent work

Insights, strategy candidates, tool calls, validation results, and operator actions can be inspected after the fact.

Provider flexibility

Use the model providers your governance allows. The platform is built to route work through approved keys and tools.

Deployment

Built for a real operations path

TrustyGenius can start small for evaluation and graduate into a managed production footprint. The default path is Docker-based, with room to adapt to the container platform and controls your team already uses.

Evaluate

Seed data, paper account, and a focused instrument universe.

Operate

Scheduled Geniuses, dashboards, reconciliation, alerts, and audit logs.

Extend

Custom feeds, brokers, models, tools, asset classes, and Geniuses.

The point of the architecture

Low latency matters, but the larger value is discipline. The system gives autonomous agents enough context to be useful, enough controls to be reviewable, and enough memory to improve from the team's feedback instead of starting over every morning.