Bot Control Console

A serverless mobile web app on AWS to see and stop every trading bot I run, with Cognito sign-in and MFA, least-privilege IAM, and an audit trail.

Bot Control Console

Several of my projects run unattended on AWS: paper-trading bots on weekday schedules, a crypto bot on its own cadence, and more on the way. Each one has its own stop switch, but finding and flipping the right one in the AWS console from a phone is slow. So I built one page that knows all of them.

It is a mobile-first web app served from my own subdomain. One card per bot shows whether it is running or halted and when it last ran, with Halt, Halt buys, Resume and Report buttons, plus a single Halt all bots button.

The real page, rendered locally against mock API responses so no account data appears. Left: every bot running. Middle: right after Halt all bots, with the server's reply. Right: the signed-out view.

Architecture

  • Front end: a single HTML page with plain JavaScript. It signs in through the Amazon Cognito hosted UI using OAuth2 with PKCE and an authenticator-app second factor, then calls a small REST API with the access token.
  • API: an API Gateway HTTP API with a JWT authorizer, so unauthenticated requests never reach my code. Throttling and a small reserved concurrency cap what a flood of requests could cost.
  • Back end: one Python Lambda. Each bot is an entry in a registry file that names its halt switch (an SSM parameter or an S3 key), where its status lives, and how to request a report. Adding a bot means adding a registry entry and its permissions.
  • Infrastructure as code: the whole stack is a CloudFormation template with deploy scripts, and the test suite runs in GitHub Actions on every push, with no AWS account needed.

Designed for the bad day

The console exists for the moment something goes wrong, so the design choices lean toward safety:

  • Stopping is easy, and starting again is not. Halting takes one tap. Resuming also needs a fresh code from an authenticator app, so someone who got into my email or chat could stop trading but could not restart it.
  • It never holds broker keys. Its IAM role names exactly the switches, status files and report functions it uses. It cannot read the trading API keys and cannot place an order.
  • Halt means “no new orders,” never “sell everything.” A forced sale at the bottom of a bad day is its own risk.
  • A remote control, not the engine. Each bot checks its own switch before every run. If the console breaks, the bots carry on, and every bot can still be stopped the old way.
  • Every command is logged to CloudWatch for 90 days. Alert emails carry a halt link that opens a confirm button, so a mail scanner opening the link changes nothing.

Stack

JavaScript, Python, AWS Lambda, API Gateway (HTTP API), Amazon Cognito, SSM Parameter Store, S3, CloudWatch, CloudFormation, GitHub Actions. Built with Claude Code.

The source is in a private repository, because it is wired to my own accounts. I’m happy to walk through it on a call.

Related: Serverless trading bots on AWS, the bots this console controls.