[e925]
░ armory · agentic-autonomy · compare

browser-harness vs karpathy/autoresearch

Both in the agentic & browser category. Side-by-side — pick the one that fits your stack tonight.

browser-harness★★★★★
✓ loya-tested🆓 free🔌 drop-in

The fastest way to let Claude drive your browser. Self-healing, works on any site.

rating
5
tested
✓ loya-tested
cost
free
install
drop-in
stars
4,697
updated
5d ago
#claude-code#browser#chrome#automation#python#open-source
avoid if

You don't need your agent to control a real browser.

open the full entry →
karpathy/autoresearch★★★★
🆓 free🛠️ wire-up

Karpathy's proof-of-concept for letting AI agents do research on their own.

rating
4
tested
cost
free
install
needs-wiring
stars
75,483
updated
4w ago
#karpathy#agent#research#autonomous#python#case-study
avoid if

You want a ready-to-use tool. This is research — better studied than installed.

open the full entry →

why it matters · browser-harness

Need Claude to log into something, scrape a page behind a paywall, fill out a form, or click through a real app? browser-harness spins up a real Chrome that Claude can control — screenshots, clicks, typing, iframes, popups, and it restarts itself if anything breaks. If you want your agent to actually do things on the web instead of just talking about them, this is the one. Free, open source.

why it matters · karpathy/autoresearch

Andrej Karpathy built this to show what it looks like when you point AI agents at a scientific problem and let them iterate — they propose experiments, run them, read the results, propose new ones. Minimal babysitting. Not a drop-in tool you install. It's a pattern to learn from: how to scope a goal, what a feedback loop looks like, where to add guardrails. If you're trying to build an autonomous agent for your own topic, study this one as a case study.

more agentic & browser to compare

derived live from the armory manifest · same-category only