How It Works

F7 uses a local-first architecture designed around a core principle: process data on the device, transmit only what's needed, and protect privacy at every step.

Data Flow

┌─────────────────────────────────────┐
│          Employee's Device          │
│                                     │
│  Work activity → Privacy filter →   │
│  Local AI model → Encrypted store   │
│                                     │
│  ✓ Only metadata (app names,        │
│    timing, click counts)            │
│  ✗ Never: prompts, files, emails,   │
│    screenshots, clipboard           │
└───────────────┬─────────────────────┘
                │  Encrypted (TLS 1.3)
                │  Structured metadata only
                ▼
┌─────────────────────────────────────┐
│           F7 Controller             │
│                                     │
│  Receives metadata → Computes       │◄──────────────────────┐
│  scores → Powers dashboards         │                       │
│                                     │   ┌───────────────────┴───────┐
│  ✓ Tenant isolation per org         │   │  Third-Party APIs (opt-in)│
│  ✓ Role-based access control        │   │                           │
│  ✓ Comprehensive audit logging      │   │  ChatGPT, Microsoft 365,  │
└───────────────┬─────────────────────┘   │  GitHub Copilot, etc.     │
                │                         │                           │
                │                         │  ✓ Usage metadata only    │
                │                         │  ✗ Never content          │
                │                         └───────────────────────────┘
                ▼
┌─────────────────────────────────────┐
│           Dashboards                │
│                                     │
│  Executives: team & org analytics   │
│  Managers: team-level insights      │
│  Employees: personal data only      │
└─────────────────────────────────────┘

Step 1: On-Device Observation

The F7 agent runs on each employee's device and observes work patterns:

• Which applications are in use (app name and category — never window content)
• Activity levels (click counts, keystroke counts — never individual keystrokes)
• AI tool interactions (which AI provider, turn count, response sizes — never prompt or response text)
• Session structure (focus time, context switches, session duration)

The agent includes a privacy filter that strips personally identifiable information before any further processing.

Step 2: Local AI Classification

A purpose-built on-device AI model runs entirely on the device to classify work patterns:

• Categorizes sessions by type (deep work, collaboration, admin, etc.)
• Scores AI interaction depth (surface use vs. integrated workflow)
• Detects anomalies locally

This classification happens before any data leaves the device. The model never sends prompts, responses, or content to any server.

Step 3: Secure Transmission

Only structured, classified metadata is transmitted to the F7 Controller:

• Encrypted with TLS 1.3 in transit
• Authenticated with per-device cryptographic credentials
• Compressed using Protocol Buffers for minimal bandwidth

What's transmitted is a structured record — app names, timing, counts, and classification labels. Never raw content.

Third-Party Integrations (Opt-In)

Organizations can optionally connect F7 to third-party apps (e.g., ChatGPT, Microsoft 365, GitHub Copilot, Grammarly) via their APIs. When enabled, F7 retrieves usage metadata — session counts, feature adoption, seat utilization — never document contents, message text, or prompts. Each integration must be explicitly authorized by an administrator.

Step 4: Scoring & Analytics

The F7 Controller computes insights from all data sources — agent metadata, employer-provided HR data, and third-party integration data:

• AIQ Score: A composite measure of AI adoption sophistication
• Workflow patterns: How teams integrate AI into their work
• Trend analysis: How adoption changes over time

Step 5: Dashboard Access

Insights are presented through role-appropriate dashboards:

Role	Sees	Access Method
Executive	Org-wide and team aggregates	Authenticated web dashboard
Manager	Their team's analytics	Authenticated web dashboard
Employee	Only their own data	Personal dashboard (planned)

Managers see team-level patterns. They do not see individual employees' raw activity — only aggregate insights and scores.

---

Key Takeaway

The F7 agent does the heavy lifting on the device. By the time data reaches the server, it's already structured metadata — no content, no PII, no surprises.