ArgusArgus

Powered by MiroMind

Deep research, only where it counts.

Verifying a factual claim isn't a database lookup — it's research. Argus hands each one to MiroMind's mirothinker-1-7-deepresearch, an agentic model that searches the live web, opens sources, and reasons its way to a verdict — then shows its work.

Tools the model calls on its own

thinkingweb_searchfetch_url_contentexecute_python

Autonomous, per claim

The model decides how to investigate each claim — which queries to run, which pages to open, when to compute. In the legal demo, the verifier records 76 web searches across 5 selected claims before producing its verdicts. No hard-coded search path.

A trace you can trust

Every search, fetch, and reasoning event is streamed and recorded as it happens. A stateful SSE decoder stitches events back together across network packets, so the saved trace reflects what the model did — no dropped steps, no mangled source URLs.

Parallel, idempotent, capped

Claims run concurrently, each as its own background job. Deterministic idempotency keys make retries safe, and a hard budget cap stops a long audit before spend can run away.

Reserved for the verdict

Only verification calls MiroMind. Parsing, claim extraction, the consistency check, and report writing run on a cheaper model — so deep research is spent on the answer, never the plumbing.

See the reasoning, end to end.

Every search, fetch, and step is recorded and replayable in the audit cockpit — grouped by claim, down to each source link.