Architecture

Agent-first without black-box computation

The story is how agents, internal data, deterministic engines, and human approval connect into one operating layer.

Agent Interface

Research console, run records, approvals, and reporting entry points.

Task planning, tool selection, state management, retries, and handoffs.

Notebooks, Python, backtest engines, schedulers, alerts, and reporting tools.

Internal research data, quality summaries, source links, and version records.

Reproducible computation for backtests, risk, statistics, and portfolio constraints.

Permissions, human approvals, audit logs, and paper/live separation.

Integration surface

DQT can connect to research environments while respecting tool, data, and approval permissions.