Trace — Record, Sketch, Diagnose

Trace is a local-first screen recorder for people who need to capture, mark up, and explain what happened on a web page — without sending a single byte to a server. No signup. Runs locally. Forever free. RECORD • Capture the current tab, a window, or your whole screen. • Optional microphone and a draggable camera bubble. • Two clicks and you're recording. Crash-safe: footage is written to disk in 5-second chunks, so a browser crash loses at most 5 seconds. SKETCH (Draw mode) • A floating control Pill unfolds a Paint Tray over the page while you record. • Calligraphic pen whose stroke width responds to speed and stylus pressure, plus highlighter, arrow, and spotlight tools. • Annotations are stored as vector data with timestamps — they can fade automatically and replay themselves (draw-on animation) live and in the exported video. • Click-anticipation rings draw the viewer's eye to where you're about to click. DIAGNOSE • Capture the page's console output and network requests alongside the video (opt-in per recording). • On-device subtitles via Whisper (downloaded on first use, runs fully offline afterward). • AI Check turns the recording's console, network, and action log into a plain-language diagnosis using any OpenAI-compatible endpoint you configure — including local servers like Ollama or LM Studio, for analysis with zero cloud calls. EXPORT • WebM (instant), MP4 (H.264), or looping GIF — sized to drop straight into a Jira ticket. • Trim, split into clips, redact/blur regions, and burn in subtitles. PRIVATE BY DESIGN • No backend, no account, no telemetry. • Nothing leaves your device unless YOU explicitly invoke a network feature (the Whisper model download, or an AI Check against an endpoint you chose). Each of those shows its opt-in state before it runs. • The action logger records element labels and roles — never what you type into fields. Password and payment fields are never logged at all. Trace is open and inspectable. If you can record it, you can keep it.