Browser agents for commerce work with visible evidence.
Layer 8 turns manual page checking into repeatable browser work. Start with Korean price comparison, then reuse the same evidence layer for marketplace sellers, reviews, social signals, and browser QA.
Layer 8 sits above the classic OSI stack: the layer where software understands a human business task, uses browser surfaces, and keeps the work visible.
The product is an operator surface for real web work.
Every job starts as a function applied to a target: check availability, price, reputation, or compliance across stores, ecommerce platforms, social surfaces, and suppliers.
Layer 8 watches a target with a function: availability, price, reputation, or compliance across stores, platforms, social surfaces, and suppliers.
Control plane above worker nodes.
Layer 8 separates the control plane from worker nodes: the control plane defines, stores, and reviews the job; worker nodes execute the browser work.
flowchart TB
subgraph ControlPlane["Control plane"]
direction TB
Dashboard["Dashboard
task setup, review surface"]
Store[("Database
results, screenshots, routines")]
Runner["Routine checker
cron, schedule, on-demand runs"]
Router["Browser routing
preferred tool and fallback chain"]
end
subgraph Workers["Worker nodes"]
direction TB
PW["Playwright
default worker"]
CF["Camoufox
same-network worker"]
LP["Lightpanda / CDP
same-network worker"]
EXT["External browser services
Browserbase, Cloudflare, browser APIs"]
end
Dashboard --> Store
Store --> Runner
Runner --> Router
Router --> PW
Router --> CF
Router --> LP
Router --> EXT
The dashboard is where work is configured and reviewed.
Runs can start on demand or from a schedule. A first run can browse normally; repeated work can use memorized paths so the next run reaches the right page with fewer steps.
- Simple browsing: resolve target, open a browser tool, observe page, act, capture result.
- Optimized replay: reuse memorized routines for known sites, forms, filters, and result pages.
- Storage: keep results, routines, screenshots, source links, and run notes.
- Output: show findings in the dashboard now; email notification can be added later.
Worker nodes can be local, internal, or external.
Each task can use a default Playwright worker node, internal Camoufox or Lightpanda worker nodes on the same network, or external worker nodes when a target needs a different runtime.
- Configurable routing: a task chooses its preferred browser tool and fallback chain.
- Network-local worker nodes: Camoufox and Lightpanda can run beside the control plane.
- Provider flexibility: Browserbase, Cloudflare, or other browser services can sit behind the same interface.
A run starts from a trigger and ends as evidence.
Layer 8 keeps the run loop visible: resolve the right surface, act through the selected browser tool, prove the result, and repeat with memory.
flowchart LR Trigger["Cron / schedule
On demand"] --> Dashboard["Layer 8 dashboard
configure and review"] Dashboard --> Store[("Database
results, screenshots, routines")] Dashboard --> Loop["Internal run loop
resolve, browse, observe, act"] Store --> Loop Loop --> Router["Browser tool router
preferred tool plus fallback chain"] Router --> Playwright["Default Playwright"] Router --> Internal["Same-network workers
Camoufox, Lightpanda"] Router --> External["External browser services
Browserbase, Cloudflare, browser APIs"] Playwright --> Capture["Captured result
fields, screenshots, source links"] Internal --> Capture External --> Capture Capture --> Store Capture --> Notify["Dashboard result now
email notification later"] Notify --> Dashboard