WinLayer Docs
Product documentation · Beta behavior. A documented path is not production acceptance; availability still depends on source health, workspace plan, provider setup, and deployed evidence.

AI briefs & grounding

Every agent — capture brief, recompete brief, proposal prep, executive analyst, and Daily Growth Copilot — runs on grounded workspace evidence.

  • The agent receives a numbered evidence pack — the only facts it may use.
  • Every claim that rests on a fact must cite its evidence id inline, like [E3]. Citations are hoverable in the UI.
  • Citations to evidence that doesn't exist are flagged as ungrounded — no silent hallucinations.
  • Low-confidence or thin-evidence outputs carry a review recommended flag for a human.
  • Every run logs model, tokens, latency, and cost to the spend dashboard.
  • The Daily Growth Copilot can propose a pursuit-status action, but it remains pending until an authorized human approves or rejects it; the decision and any execution are audited.
Thin evidence lowers confidence — it is never padded over. If the pack doesn't contain a fact (say, a contract end date), the brief says so plainly instead of guessing.