This one’s ai-slop, as advertised on the about page — built with Claude Code over a couple of months of on-and-off sessions.

Most people who read a lot online never see any of it again. Someone opens an article, skims it, closes the tab, and forgets it — buried inside browser history nobody searches. Archivore, a small command-line tool, exists to fix that one habit at a time.

The idea is simple: capture everything worth keeping, automatically. Archivore scans browser history since its last run, picks out links from Hacker News, Reddit, and X, downloads the linked article, and converts it to Markdown. Nothing requires manual saving or tagging. Opening a link in a browser is enough.

What happens to that Markdown is the useful part. Every file lands in an Obsidian vault with proper front matter — title, source, author, the date it was actually read — so a capture becomes part of a personal wiki rather than an orphaned text file. An index tracks everything gathered, and a semantic-search tool called qmd indexes the whole vault, so old reading can be found by meaning, not just keyword.

Running archivore on more than one machine used to mean duplicate downloads, since each kept its own local queue. A small, free-tier Cloudflare Worker backed by a D1 database fixed that: every machine now claims and completes items against one shared queue, so nothing gets fetched twice. Four network calls per run keep every machine in sync, regardless of how many articles were captured.

Reduced to its essentials, this is what running it on several machines looks like: each one scans its own browser history independently, but they all claim and complete against the same queue and write into the same vault, so the result is one deduplicated wiki no matter how many machines fed it.

flowchart TD
  M1["Machine 1
scans browser history"] <--> Q[archivore-queue D1]
  M2["Machine 2
scans browser history"] <--> Q
  M3["Machine 3
scans browser history"] <--> Q
  M1 --> V[shared vault]
  M2 --> V
  M3 --> V
  V --> S[qmd search]

None of it publishes anything yet. Archivore’s second half — turning weeks of captured reading into actual essays, using Claude — stays on the roadmap. For now it does the unglamorous part well: it remembers, so its owner doesn’t have to.

Session stats

The session behind this post ran from July 1st to August 31st, 2026 — two months, on and off, in one continuous conversation. Across 1,779 assistant turns, it used 3,402 fresh input tokens, 1,632,727 output tokens, 22,040,856 cache-creation tokens, and 661,622,906 cache-read tokens. That last figure looks alarming until you account for pricing: cache reads cost a fraction of fresh input, and most of that total is Claude re-reading context it had already built up, not new work.