In one sentence
Ask the Library is a chat-style search over your document register — type a question in plain English, the agent reads your controlled documents (internal and external), and returns an answer with citations pointing at the specific sections that support it.
- It only reads what you can read. Search results, citations, and answers are filtered by your document access — the same permission rules that govern the rest of the app. Two users asking the same question can legitimately get different answers if their access differs.
- Every answer carries citations. Each citation links to a specific heading in a specific approved document version, so you can verify the answer rather than trusting the agent's paraphrase. If the underlying document changes later, citations get a staleness badge so you don't unknowingly act on outdated text.
- A conversation can be attached to a record. Linking a chat to an NCR, MOC, audit, or other record freezes the conversation as part of that record's evidence trail — useful for audit findings and investigation history.
Where it lives
Sidebar → Ask, available to anyone in your org with read access to documents.
The page has two panes — a conversation list on the left, the chat thread on the right. Each thread is preserved indefinitely (until you delete it), so coming back to "that question I asked about lockout/tagout last Tuesday" is one click.
What the agent searches
The agent reads two kinds of source:
- Internal documents — your TipTap-authored SOPs, policies, procedures. Indexed by section heading, so citations land on a specific heading inside the document.
- External documents — uploaded PDFs (ISO standards, customer specs, supplier docs). Indexed page-by-page during ingest, so citations land on a page number.
In both cases the index is built from approved versions only. A draft revision sitting under review doesn't change what the agent sees until it's approved. Once approved, the index is refreshed and the new content becomes the cited source.
Scoping a question
Two scopes narrow what the agent looks at, both pre-set before you type the question:
- Active facility. If you have a facility selected in your top-bar scope picker, the agent only searches documents scoped to that facility (plus org-wide docs). Switch your active facility to All facilities to broaden the search. The page shows the current facility under the input as a reminder.
- Departments. A multi-select picker above the input lets you restrict the search to specific departments. Leave it empty (the default) to search across every department in the current facility scope.
Both are interaction polish, not security — your underlying document access is the hard boundary. Scoping makes a "give me the chemical-handling rules for the Phoenix plant" search productive instead of returning corporate-wide noise.
Citations and verification
Every answer turn lists the citations that produced it. Each citation includes:
- The document number and title, linking back to the document.
- The section heading (for internal docs) or page number (for external PDFs).
- An optional snippet showing the cited text.
Click a citation to open the document at the cited section. The agent never invents source material — if it can't find a defensible citation, it tells you it doesn't have enough information rather than hallucinating an answer.
Citation staleness
Documents change. An NCR investigator might come back to a six-month-old Ask conversation that cited a procedure that's since been revised. The agent re-checks every citation when you re-open a conversation and flags four kinds of drift:
| State | Meaning |
|---|---|
| Updated since | A newer approved version of the cited document exists. Your citation points at an older version — accurate to what was true at answer time, but no longer the current text. |
| Section moved | The heading the citation pointed at no longer exists in the current version. The section was renamed, removed, or restructured. |
| Withdrawn | The cited document is now marked Obsolete. |
| Access revoked | You no longer have read access to the cited document. The document may still exist; you just can't read it now. |
Stale citations show an inline badge in the conversation view. The original answer is preserved exactly as it was — Ask doesn't rewrite history — but you get a clear signal that the underlying source has drifted, so you can re-ask if needed.
Attaching a conversation to a record
A typical Ask flow ends with "OK, that's the answer I needed for this NCR" — and you want the chat preserved as evidence of how you reasoned the response. Click Attach to record on the conversation, pick the target (NCR / MOC / Audit / Meeting / Risk / FMEA / JHA / Exemption / Document), and the conversation becomes linked.
Two things happen the moment a conversation is linked, even once:
- The conversation freezes. You can re-open it and reference it, but you can no longer add new turns. The record-linkage commitment means the chat can't drift after the fact.
- It surfaces on the host record's detail page. A side panel on the linked record lists every conversation attached to it, viewable by anyone with access to the record. The conversation owner is shown alongside; this is access-widening by design — colleagues working the same NCR see your reasoning.
Unlink later if you want, but the freeze stays: once any link existed, the conversation is permanently read-only. This is intentional — a regulator asking "why was this conversation evidence, and why did it change after the linkage was removed?" shouldn't have a story to write.
Conversations and history
Every prompt-and-answer is stored as a turn on a conversation. The left pane lists every conversation you own, newest activity first. Click any one to reload it.
Conversations are owner-only by default — your colleagues can't see your unlinked threads. Linking widens access to people who can read the host record, which is the only access-widening path. The list filters apply automatically: facility-scoped users still see all their own threads, since ownership trumps scope here.
Conversation deletion is a permanent action — the linked-record copies of the chat go away with it, since the source is your conversation. Linked conversations show a warning at deletion time.
When AI is off
Ask the Library is part of the AI feature set:
- If your org has AI turned off at the org level (Admin → Organization → AI Settings), the page renders a notice rather than the chat UI. No queries leave the system.
- If AI is on but the Ask-the-Library agent is disabled specifically (one of the per-feature toggles), the page shows a similar notice.
Both states are admin-controlled. See AI Features & Settings and AI: BYOK vs Managed for what's billed and how providers are picked.
What it doesn't do
- Ask doesn't write. It only reads documents and answers questions. It doesn't create NCRs, raise actions, or edit documents — those are separate flows.
- No cross-org search. The agent strictly stays within your organization's documents. Even managed-AI deployments don't share document indexes across customers.
- No drafts. The index is built from approved versions; an SOP sitting in Draft hasn't been indexed and won't be cited.
- No image OCR. External PDFs are text-extracted; images embedded in PDFs aren't OCR'd today. If a critical clause exists only as a scanned image in an external doc, the agent will probably miss it. The internal-document side is fully text-based, so this only affects scanned externals.
Related
- AI Features & Settings
- AI: BYOK vs Managed
- Documents
- External documents
- Document downloads — the access trail also captures Ask-the-Library reads
- Read-and-acknowledge users