Internal knowledge first, then specialists

The idea

A worker needs internal evidence before its first web search or question: a knowledge or memory read, a colleague's answer, or a recorded miss. After that it can consult read-only specialist sub-runs, search the web and, with approval, ask an outside expert.

The problem
Workers searched the web or asked people before they used what the organization knows.
Why it matters
The organization's own knowledge comes first, and outside answers are cited evidence, not instructions.
Where it stands
In production as of

A step-by-step replay of this idea is not published yet. The library adds one when a recorded run shows it.

  • Employee supervises its agents

    The AI employee plans, briefs, reviews, integrates and talks to people. It hands bounded plan slices to cheaper models or agent CLIs.

  • Define deeply before building

    Before building, the worker defines what, why, for whom and how. It checks fit, knowledge and authority, and answers its own questions with research and with colleagues.

  • Review threads and CI on the exact head

    Each actionable PR comment becomes a plan step that stays open until answered. Red CI on the exact head sends the work back. A merge is compared with what was verified.

  • Evidence-gated questions and settled decisions

    A blocking question is refused until research cites a tool result, and asking again about a settled decision needs new evidence. A decision that arrives after a completion claim sends the work back.

  • Hybrid memory recall

    Recall ranks memories with full-text and trigram search, plus vectors where embeddings exist. It says which arms ran and why vectors are off, and it gives the worker its readable memory as files.

  • Typed requests in every direction

    Workers and people ask each other typed questions (review, information, decision, help, handoff) peer to peer, up or down, with due times, defaults, redirects and an inbox.

Turn the numbers into shipped work.

Crewshal runs these choices for you: a persistent AI worker with memory and rules, on your Claude and Codex subscriptions.