Running the simulator in the browser keeps the SaaS setup simple (no installation), while still allowing strategies to run in a fully unrestricted local environment.

I started working on this while experimenting with AI coding agents writing trading strategies. Most backtesting platforms execute strategies inside their own infrastructure, which usually means restrictions on compute, libraries, GPUs, or external APIs.

This system separates the simulator and the strategy runtime.

Architecture roughly looks like this:

Browser ↓ WebAssembly market simulator ↓ WebSocket ↓ Local Python strategy

Because strategies run locally, the strategy environment is effectively unrestricted. You can use GPUs, external APIs, local datasets, or basically any Python library.

This setup also works well with AI coding agents (like Claude Code). An agent can write strategies, run backtests, analyze results, and iterate.

The platform also exposes APIs designed for autonomous agent workflows where an AI agent can run strategies, evaluate results, and improve them in a loop.

On the analysis side, positions are traced at the ticket level rather than immediately aggregated, so the lifecycle of each ticket can be inspected in the UI. There are also diagnostics like exit analysis, MFE/MAE statistics, equity curves, and drawdown analysis.

The backend is a lightweight Django service running on AWS.

It's still early and probably rough in places, but curious what people think.

It's currently free during preview while I gather feedback.

App: https://app.hawk-backtester.com

Python client library: https://github.com/nima555/hawk-bt