The promise you are making
When you give an agent a knowledge base, you are telling it "act on what you find here." That promise is only safe if a person looked at every fact before the agent did. The distiller is the part of Ship that holds the line. It takes raw text from your sources — a docs repo file, a web page that changed overnight, an upload, a note routed in from a chat — and turns it into a draft article, then stops. A human opens the draft, reads it against the source, and decides what to do.
Nothing becomes knowledge silently. That is the entire design.
How a draft is born
A source refreshes. Something changed. The distiller reads the new text and does three small jobs.
It picks a topic. Most documents are about one thing — the deploy procedure, the lint rules, the customer-tier ladder. The system reads the document and names what it is about, then matches that topic against the buckets you already have. If the match is clear, it nominates a bucket. If no existing bucket fits, it flags the draft as needs routing and waits — it will not invent a new bucket on its own.
It writes a short article. A clear title. The fact as the source states it. An example only where the example removes ambiguity. A provenance footer that links back to the source so any reader, human or agent, can walk the pointer.
It stops. The draft is private. No agent sees it. No citation can land on it. It lives in your review queue until you, or a teammate, decides what to do.
The four review actions
Open a draft and you have four moves.
- Publish. The draft becomes a live article. The next agent that reads the bucket picks it up. Citations begin to accumulate. This is the everyday action — the article looks right, you adjusted a word here and there, you publish.
- Supersede. An older article on the same topic already exists, and this draft is its replacement. The system links the two together. The old article stays readable so the history is clean — anyone who needs to know why we changed our mind can read both versions. The new one becomes the active one.
- Archive. The fact stopped being true. The vendor changed their API, the procedure was retired, the rule was relaxed. Archive keeps the article and its history readable but removes it from agent searches. The archive is your memory of decisions; it is not a delete.
- Send back. The draft is incomplete or unclear. The source did not give the distiller enough to work with. Send it back and the system marks it as needing more context. Pair it with a new source, edit it directly, or leave it pending until your team can answer the question it raises.
You can also escalate a draft to the Inbox. If the article touches policy and one reviewer is not enough — the security stance, the brand voice, the legal disclaimer — send it through the same approval flow your team uses for any other decision that needs more than one set of eyes.
What you see in the queue
The review queue is a filterable list. You can scope it to one bucket, to one source, to drafts created since a particular date, or to drafts that have been sitting more than a few days. Most teams find the queue is healthiest when reviewers spend ten to fifteen minutes on it daily — long enough to clear what came in overnight, short enough to fit between meetings.
The queue tells you something honest about your setup that no dashboard would. If it is growing, one of three things is true. Your sources are too aggressive — you are importing noise that does not belong. Or your buckets are too broad and need a split because reviewers cannot tell what goes where. Or your team is routing chat notes that should have been answered and forgotten instead of becoming articles. Read the queue as a diagnostic. A calm queue means your knowledge base is in tune. A backed-up queue means a tuning conversation is overdue.
If the queue is empty, you are probably not importing enough — or your team is answering many questions that nobody is writing down.
Why this shape
Other systems will offer to "auto-publish high-confidence imports." We do not. Auto-publish is how a vendor doc page change at 2 a.m. becomes Tuesday morning's wrong agent behaviour — with no audit trail of who decided the change was an upgrade. A reviewer who spends thirty seconds on a draft has done more for trust than any confidence score the model could attach to its own work.
The review path is the cost of trusting your agents with knowledge. The cost is small. Ten minutes a day, divided across the team. What you buy with it is the line between "we collected some text" and "agents now act on it." That line is worth holding.