How Vantic IQ works

​

The signal, not the content

Vantic IQ derives all of its intelligence from workflow signals — structural facts about how work moves on your Mac. It does not read what applications contain. Everything in this page describes what the product is doing at each stage, from first launch to Operating Assessment.

---

​

Five layers

Vantic IQ is built in five distinct layers. Each layer has a defined scope and does not reach into the layer above it for signal data.

​

1. Account and profile

Handles sign-in, onboarding, and the profile your conversational interface draws on.

• Sign-in: Email and password or magic link via macOS app.
• Onboarding: Six questions — name, business name, sector, team size, primary friction area (optional), and average team hourly rate (optional). Takes approximately 60 seconds.
• Profile storage: Stored in Supabase on AWS ap-southeast-2 (Sydney). Workflow signal data is never stored at this layer.
• Hourly rate: Used only by the ROI calculation in the Operating Assessment. Not used for pattern detection.

​

2. Signal detection

Detects workflow signals from your Mac and prepares them for the Pattern Engine.

• Source: macOS Accessibility API. One system permission grant.
• What is captured: Active application (bundle ID), event timestamps, session duration per application, transition sequence and timing.
• Content-adjacent fields: Window titles and filenames are SHA-256 hashed immediately at the point of detection. The original value is discarded and cannot be recovered. Hashes are used only to detect repetition and frequency.
• Exclusion engine: Runs before any event reaches storage. Password managers, Australian banking applications, and system credential prompts are permanently excluded and cannot be re-included.
• Audit log: Every event generates a verifiable audit entry confirming that the exclusion check and hashing were applied.

​

3. Pattern Engine

Turns workflow signals into friction scores. Runs locally on your device. Four dimensions are calculated independently: Each dimension produces an internal score used by the Pattern Engine only. The output your conversational interface and reports receive is the Friction Score (0–100) — the customer-facing composite. Internal engineering scores are never surfaced to customers. The Pattern Engine applies a confidence gate before producing any result. See The confidence model below.

​

4. Conversational interface

Turns pattern data and your profile into answers, insights, and suggested next steps.

• Powered by the Claude API (Anthropic). Raw workflow signals are not sent to the AI provider — only structured, derived pattern context.
• Before a Friction Score is ready: the interface draws on your sector knowledge and User Context Hints. Friction Pulse signals (single factual signals) appear in the first 1–4 days.
• After a Friction Score is published: grounded in your named workflow patterns. Can explain dimension scores, trends, and specific workflows by name.
• Conversational input is session-scoped. It is never persisted between sessions and does not feed the Pattern Engine.

​

5. App Shell

The menu bar interface, Privacy Hub, settings, and Operating Assessment export.

• Menu bar: Active (solid), Paused (dimmed), No data yet (subtle pulse), Suggestion available (badge), Report ready (badge).
• Privacy Hub: Application exclusions, time window exclusions, purge controls.
• Operating Assessment export: Renders to a self-contained HTML file opened in your default browser. From there: File → Print → Save as PDF.
• Billable Hour Map: Available on all tiers. On Diagnostic: ROI-anchored view tied to the day-14 Operating Assessment. On Starter and Growth: continuously updated, including deviation tracking and AI-tool metrics when entitled.
• Starter / Growth continuous surfaces: Weekly Friction Brief, Deviation Banner, Monday Morning Operating Leak, optional Pattern Labels, and AI-visibility surfaces. Growth adds the Governance Receipt. See Glossary and AI visibility.

---

​

What workflow signals are

Workflow signals are the only data Vantic IQ uses to derive patterns: Together these describe the structure of how work moves — not what the work contains.

---

​

What is never captured

• Screen content, screenshots, or any recording of your display
• Keystrokes or typed text
• File or document contents
• Email bodies or message text
• Clipboard contents
• Window title text (hashed at detection; original discarded)

This is a structural constraint, not a setting. There is no code path that reaches these data classes.

---

​

The confidence model

Vantic IQ uses a two-tier confidence gate before surfacing results.

​

Engine gate (minimum threshold)

The Pattern Engine requires sufficient signal before producing a result. The minimum confidence threshold requires:

• At least 2 distinct sessions detected
• At least 10 workflow signal events in the evaluation window
• Sufficient calendar coverage across the evaluation window

Below this threshold, the Pattern Engine returns nothing. No score, no estimate, no partial composite.

​

Display gate (full confidence threshold)

The composite Friction Score is published in the UI only when aggregate confidence meets the full display threshold. This is a higher bar than the engine gate. Between the engine gate and the display gate, the conversational interface may surface partial pattern context — without claiming per-dimension readiness — while the composite score waits for stronger signal. The display gate is deliberately the higher bar: the headline score appears only once the signal genuinely supports it. Why this matters: A score produced from thin data would be misleading. If your first week was unusually busy, quiet, or non-representative, the Pattern Engine waits for a clearer picture. This is by design. The evaluation window is 14 days by default. First-insight evaluation uses a 7-day window at the same engine gate.

---

​

Local vs cloud

---