Ask Alex
anything.
Alex is the conversational layer over QuantumEye. The ninth module, and the one that knows the other eight. Ask the day's questions in plain English. Talk a new automation rule into existence. Alex drafts it, you confirm in the dashboard. Role-aware. Audit-logged. Built on real platform APIs.
search_eventsget_business_intelligenceAsk in plain English. Get an answer, not a search.
Alex turns a question into the right tool call against your live data, then narrates the result, with every action logged.
Ask Alex
AINatural-language search across your estate
Quiet night. The After-Hours Guard automation stayed armed from 22:00 and nothing tripped a watchlist or restricted-zone rule. The only movement was the 04:00 goods-in delivery on CAM-02, which was whitelisted, so no alert was raised.
16 tools, live. Read, draft, and act, with confirmation.
Alex doesn't make answers up. Every response is a call against a real QuantumEye API, search_events, get_individual_profile, draft_rule, update_event_status, approve_theft_for_report and the rest. If a tool returns nothing, Alex says so.
Search & investigate events
What happened overnight? Show pending theft events. What's the audit chain on this incident? Cross-referenced with your face graph and prior events.
Lookup individuals & watchlists
Tell me everything about person X. Who's whitelisted in the Suppliers group? Is this face on the watchlist? Surfaces profile, history, and last-seen across the estate.
Business intelligence
Footfall today vs yesterday. False-alarm rate this week. Detection accuracy week-over-week. The morning brief turned into a chat.
Talk a rule into existence
"Set up a PPE alert for the stockroom." "Alert me if more than 8 people queue at the kiosk." Alex drafts a ZoneRule from your sentence, polygon, type, threshold. You confirm in the dashboard before it goes live.
Confirm, reject, approve
Confirm a theft or mark a false alarm. Approve a confirmed theft for the police report. Toggle a rule on or off. Every action requires inline confirmation and is server-side RBAC-gated.
Reports & notes
Draft the police report for the confirmed concealment, face, clip, audit chain, prior events. Pin an audit-logged note to an event or an individual.
Ops briefing. Watchlist lookup. A rule, talked into existence.
Each thread maps to a real Alex tool. The interactions below are illustrative wording, the underlying tool calls are exactly what runs in the product today.
get_business_intelligenceget_individual_profiledraft_reportdraft_ruledraft_rule30 questions mapped. 16 answered today.
This is the kind of conversation a store manager has with Alex before the doors open. Every reply is a real tool call against your tenant, never a guess. The two at the end are mapped and on the roadmap.
16 of 30 mapped questions are answered today. The rest are on a published roadmap, and the full matrix lives in your tenant.
Tool-calling. Not text generation.
Alex doesn't generate answers from a model. Alex calls real platform APIs that you can audit. The model interprets your question, picks the right tool, runs it, and narrates the result. The data is always real.
LLM picks the tool
Your question is parsed against Alex's tool catalogue (search_events, get_individual_profile, get_business_intelligence, …). The model picks the right one.
RBAC server-side
The tool runs against the same APIs your dashboard uses. Permissions enforced server-side, a Store Manager can't query another store via prompt injection.
Every action audited
Every Alex action that touches data writes to the append-only audit log, same chain of custody as a manager clicking a button in the UI.
Anthropic Claude
Sub-processor disclosed on the trust centre. Disabled per-tenant on request.
Real platform APIs
Same endpoints as the dashboard. No mock data, no LLM-generated facts.
Confirmation UX
Action-taking calls (status changes, police-report approval, rule toggles) require an inline confirmation step. Never auto-executed.
Graceful failure
If a tool errors or returns nothing, Alex narrates the failure cleanly rather than inventing an answer.
An LLM in a surveillance product needs a different bar.
Alex is held to the same GDPR commitments as the rest of QuantumEye, and a few extra ones that come with using a model.
No PII in the prompt context window
Face vectors, ethnicity codes, and raw video never enter the LLM's context. Alex sees structured tool results, not embeddings.
Human-in-the-loop for any write action
Every action-taking Alex call requires an inline confirmation. Status updates, police-report approval, rule toggles, all gated. New rules are drafted by Alex and activated by a human in the dashboard.
Sub-processor disclosed
The model provider is on our sub-processor list. Tenants can disable Alex entirely if their DPA forbids it.
Same audit chain
An Alex-initiated action is indistinguishable in the audit log from a UI-initiated one, and equally provable.
See Alex run against your real estate.
We can stand up a sandboxed Alex instance against an anonymised copy of your data, so the questions you actually ask are the questions you'll see answered on the demo call.