Watch an agent see, reason, and act on the open web.
Vision-language navigation agent v1 — a control room for browser-automation runs, every screenshot, every mark, every model decision recorded frame by frame and replayable on demand.
ENGINE gpt-4o-vision
ALLOWLIST 4 domains
ACTIVE RUNS 1
LAST STEP 340ms ago
SESSION webnav_run_0042
System status
Refreshed every 2s via HTMX pollRuns today
0+18% vs yesterday
Success rate holding at 91.4% across 4 target domains, 3 emulation profiles.
Success rate
0%
Rolling 7-day window
Avg steps / run
0
Median 11, p95 29
Active run
run_0042
step 9 · ok
Vision cost today
$0
Cassette replay saved $41.20
Recent recipes
find_and_add_to_cart.yaml · fill_support_ticket.yaml · compare_pricing_tiers.yaml
Task recipes run against the legacy portal fixture and two live allowlisted targets.