← All projects

Tab Veto'er

Working MVP · 2026

2,704 hoarded browser tabs triaged by an LLM that checks what I already own, already know, and actually need before it lets anything survive.

2,704 tabs ingestedACT/SAVE/CLOSE/DEFER verdictscanonical-URL dedup48-fixture regression suite

Every chronic tab-hoarder has the same 3,000-tab graveyard. The fix isn't willpower, it's context: a verdict engine that knows enough to say close it with receipts.

OneTab exports flow through platform-specific extractors (YouTube metadata, product pages, articles), canonical-URL normalization kills duplicate variants (tracking params, mobile hosts, URL forms), and each tab gets an LLM verdict — ACT, SAVE, CLOSE, or DEFER — grounded in cross-service evidence: the home-inventory service (already own it), the knowledge archive (already learned it), and the project tracker's vision gaps (actually need it for something real).

Hydration runs as resumable background jobs with a review dashboard for human override; a 48-fixture regression suite pins verdict behavior across prompt changes. First real-corpus sample: 83% CLOSE — the system agrees the graveyard is a graveyard, and can prove it tab by tab.

Status & limits — working MVP with real data; the long-term bet (learning my taste from override patterns) is designed but not yet earned.

Stack

PythonFastAPISQLiteOneTab ingestionplatform-specific extractorscross-service context

← All projects