Nivloo never asks you to trust a model. Each step below runs the same rules the crew runs on every research job — try them.
You confirm official public HTTPS sources and say why each one is official. Scout fetches them, never anything else, and saves what it saw.
Accepted at this step. On fetch, the server also pins DNS to public addresses, re-checks every redirect (three at most), accepts only HTML, text, Markdown or JSON, and caps size and time.
Analyst compares each new snapshot with the last one, word by word and deterministically. When nothing changed, the run ends there and no paid generation happens.
…The protocol fee defaults to 0.05% and can be changed by governance after a seven day timelock.
…The protocol fee defaults to 0.15% and can be changed by governance after a seven day timelock.
A verified fact is an exact sentence that appears in the new snapshot and not the old one. Interpretations are labeled inferences and state their limits; the rest stays unknown. An announcement does not prove a feature shipped, and source text is evidence — never instructions.
Configure the endpoint before starting.Requests time out after ten seconds.
Accepted as verified fact. It enters the brief with its evidence link.
Watchlists, runs and briefs are private to your wallet. Before a live run you accept a provider disclosure, because your focus and bounded source excerpts are sent to Orbio and its selected model provider.
Source excerpts. Scout makes a plain GET to the approved source. Only bounded excerpts around a real difference go to the model — unchanged sources are never sent.
Reading the live retention policy…
Proposed policy: seven days for raw snapshots and thirty days for results. Backup copies may remain until the disclosed backup expiry. The owner must confirm the policy before access opens.
Orbio provider documentation ↗Delete every watchlist, source, snapshot, run and brief from the active database in Settings. Cancel an active run first. Exports are plain Markdown files you keep.
Open settings ↗Research spends real money, so every run is capped before it starts and fails closed when anything is uncertain.
Live research stays off until the launch time, budgets, provider prices, retention and privacy terms are confirmed.
Each run reserves its budget atomically against per-run, daily and total caps. There are no auto top-ups.
Fetched text cannot change tools, permissions or budgets. The model gets no tools, URLs or secrets.
If new runs pause, saved results remain readable and exportable. One project per wallet during the initial window.