Obserf
Find useful places to mention your product
Quick follow-up to last week's post. Turns out the part of Obserf I use most isn't the scanning, it's sitting down with 30 findings and deciding what's worth a reply. That inbox was a pile of buttons, so 0.3 is mostly a redesign of it.
What changed:
- Each finding shows the model's verdict first, then the actual page text, then the score. "Why 55?" shows the math, because a number without its reasoning is just vibes with a decimal point.
- Everything has a keyboard shortcut. Press ? to see them all. I can get through a queue without touching the mouse now.
- Undo sits right where you made the decision, because d is one key away from s.
- If a scan finishes while you're reviewing, you get a "refresh?" banner instead of the list reshuffling under your cursor.
- Scores are plain numbers now. The green/amber colours implied a precision the ranking hasn't earned yet.
Still never posts anything. Still local, still free, still just Bun + Claude Code.
Upgrade: bun add @obserf/cli@0.3.0
New: bunx @obserf/cli init ~/dev/marketing
Honest question for anyone who's tried it: does the ranking hold up on your projects, or does the top 10 still need a lot of squinting?
I kept running into the same dumb problem with side projects: building them was the fun part. Figuring out where to mention them without becoming That Guy™ was not.
Search for a problem your product solves and you get 200 vaguely related pages. Then you spend an hour figuring out which ones are current, relevant, and actually have a reasonable place to reply.
So I built Obserf.
It searches Hacker News, Reddit, GitHub and the wider web, filters the obvious junk, then uses Claude to rank the survivors and explain why each one might be worth your time. When you find a good one, it can draft a response.
Importantly: it never posts anything. You read the thread, check the rules, edit the draft, and hit submit yourself. I have enough robots making questionable decisions on my behalf already.
It’s local-first. Your project profiles and history live in a SQLite database on your machine. Model calls go through Claude Code, so there’s no separate Anthropic API setup if you already use it. Obserf itself is free and open source; you just need Bun + Claude Code. GitHub
Getting started is basically:
bunx @obserf/cli init ~/dev/marketingcd ~/dev/marketingbun install
Then edit the generated project profile with what you’re building and the kinds of problems/conversations you want to find.
Before spending any model quota:
bun run obserf scan --dry-run
If the results look sane:
bun run obserf scanbun run obserf serve
That gives you a little local inbox for reviewing and triaging findings. GitHub
It’s early. The whole discovery → ranking → review → drafting loop works, but I’m still figuring out whether the ranking is consistently good enough to save more time than it creates. Which, conveniently, is now a measurable problem instead of a vibes problem.
Comment