Ask V5: an assistant for questions about your records.

• Reports are a bottleneck

  Every ad-hoc question queues behind BI. Simple operational questions wait days for someone to write SQL.

• Generic AI Leaks and Hallucinates

  External assistants have no access to your data and no accountability when they make up an answer.

• Role and Scope Get Lost

  A sales user shouldn't see raw deviation detail; an operator shouldn't see supplier pricing. Bolt-on tools ignore this.

• Answers with Citations

  Every answer links to the specific batch, CAPA, SOP or invoice that supports it — no unsupported claims.

• Role- and Tenant-Scoped

  Ask V5 runs inside your tenant with the caller's role — it cannot see data the user cannot see.

• Operational, Not Chatty

  Tuned for structured pulls (lists, counts, trends) and short factual answers — not open-ended prose.

• Refuses politely when uncertain

  If the data doesn't support an answer, Ask V5 says so and offers the next step, rather than guessing.

• Learns your terminology

  Aliases (part codes, plant names, campaign names) are picked up from your master data — no glossary maintenance.

Ask V5 lights up in three stages.

1. Read-only enabled. Assistant enabled tenant-wide in read-only mode with per-role scope.
2. Master-Data Aliases. Product, site, campaign aliases picked up; users test with real questions.
3. Write Actions Considered. Sites opt-in specific write actions with human confirmation.
4. Session Review. QA periodically reviews assistant sessions for utility and edge cases.

Assistants either accelerate work or introduce risk. These criteria separate a trustworthy operational assistant from a chat toy.

• Citations on Every Answer

  Does every claim link to the underlying record? Every answer carries record links.

• Role and Tenant Enforcement

  Is scope enforced server-side, in the caller's role? Scope enforced server-side; caller's role applied.

• No-Train / No-Retain Terms

  Do the underlying model providers have contractual no-train, no-retain? Lovable AI Gateway contractual no-train / no-retain.

• Refusal on Uncertainty

  Does it refuse and cite what it did find, rather than guess? Refusal is a first-class response.

• Read-only default with opt-in writes

  Are write actions opt-in per action type with human confirmation? Read-only default; opt-in writes with human confirmation and e-signature.

• Auditability of Every Session

  Is every question and answer retained for audit? Full session retention with per-message audit trail.

Ask V5 vs external chat assistants and vs BI-only tools.

Assistants over regulated data have to satisfy access-control and audit clauses.

• 21 CFR Part 11 §11.10(d)

  Assistant runs in the caller's session; role and record scoping enforced server-side.

• 21 CFR Part 11 §11.10(e)

  Human review, approval and signature actions are recorded in the audit trail through normal controlled workflows where configured.

• EU GDPR Art. 32

  No customer data leaves the tenant boundary except to the AI gateway under contractual no-retain terms.

• ISO 27001 A.9

  Same access model as the underlying V5 application.

• 21 CFR Part 11 (data access controls)
• EU GDPR Article 32 (security of processing)
• ISO 27001 A.9 (access control)
• NIST 800-53 AC-3 (access enforcement)

• Pharmaceuticals
• Medical Devices
• Food & Beverage
• Cosmetics
• Audit Trail
• Role-based access
• Reporting
• V5 AI
• Data Integrity
• Multi-Tenant

Does Ask V5 send my data to OpenAI or Google?

Queries route through the Lovable AI Gateway with contractual no-train and no-retain terms. Model providers see the prompt for the duration of the call and no longer.

Can it write to V5?

By default it is read-only. Write actions are opt-in per action type and always require a human confirmation step with e-signature where applicable.

What happens when it doesn't know?

It says so, cites what it did find, and suggests the next query — it does not fabricate.