01
Facet
Workers AI turns the question into three complementary lines of inquiry.
argus.coey.dev · Cloudflare Workflows + Workers AI + live sources
An agent that researches by assembling evidence, then growing the workflow around its gaps.
Argus runs for real: one durable workflow facets a question, searches current arXiv receipts, scores evidence with Workers AI, dispatches follow-up searches when the graph exposes a gap, then synthesizes the compact board.
Thesis. Deep research should be a dynamic workflow over a compact evidence graph, not a pile of parallel chat transcripts.
Dogfood it
Ask one research question.
Evidence board
The empty state is honest.
Run the workflow and this surface fills with real facets, source receipts, gap-triggered follow-ups, and a synthesis returned by the Workflow instance.
02
Search
The workflow fetches live arXiv results and converts receipts into bounded evidence cards.
03
Dispatch again
Low coverage or thin receipts become targeted follow-up steps, created during the run.
04
Synthesize
The final answer sees the graph, not a sprawling transcript soup.