A Flight Recorder for Web Apps That AI Can Read

What it does d3k wraps your dev server and a browser, then records the complete firehose—server stdout, browser console messages, network requests, user clicks, and automatic screenshots—into timestamped logs. It auto-detects your framework, whether you run Next.js, Django, Rails, or something more exotic, and dumps everything into a single timeline. The consumer is meant to be an AI agent: any assistant that can read a text file gets a structured black-box recording of exactly what broke and when.

The interesting bit The value is in the plumbing. By hooking the Chrome DevTools Protocol and merging browser telemetry with server logs in one place, d3k solves the context-window problem for AI debugging. It also offers a split-screen tmux mode that parks your AI agent next to the app, so it can read the log and suggest fixes while you watch.

Key highlights

• Auto-detects project type and dev server script across JavaScript, Python, Ruby, or anything with a dev command.
• Captures browser console, network traffic, user interactions, and screenshots via Chrome DevTools Protocol.
• Writes a unified timestamped log to ~/.d3k/{project}/d3k.log that any AI assistant can ingest.
• Supports headless mode for CI and split-screen tmux mode for local AI pairing.
• Includes diagnostic subcommands like errors, logs, fix, and crawl for quick human triage.

Caveats

• The README notes that split-screen mode requires tmux but does not explain how agents are installed, authenticated, or invoked beyond a command flag.

Verdict Give it a look if you already lean on AI agents for debugging and want to stop context-switching between browser DevTools, server terminals, and chat windows. Skip it if you debug without AI assistants or if your IDE already surfaces all of this context natively.