The Product
The operational memory layer your stack has always been missing
A lightweight on-premise agent. An encrypted, client-owned dataset. A reconstruction engine that converts raw event streams into hierarchical workflow representations — without any data ever leaving your infrastructure.
Capture Specification
Behavioural metadata. Never content.
The Alpyx agent captures workflow metadata — patterns, sequences, and transitions between applications — not keystroke content, not screen recordings, not document text. The distinction is architectural: the agent is not capable of capturing content.
Cross-software transitions
Window titles, active UI elements, application context, and inter-application transitions. The agent operates at the OS layer below individual apps — cross-software coverage independent of any application API.
Keyboard and mouse patterns
Keyboard event metadata (not keylogging content), mouse interaction patterns, and clipboard operation types. Behavioural signals only — never input content, never document text.
Sequence and duration
How long each step takes. Where hesitation occurs. The rhythm of multi-step workflows. Time-gap, application-switch, and semantic-context signals feed adaptive boundary detection.
Nano to mega processes
Captured events are labelled into a hierarchical taxonomy: atomic actions (seconds) → task sequences (minutes) → structured workflows (hours) → departmental flows (weeks) → organisational programmes (months).
Workflow fragments across days
Workflows often span sessions, overnight gaps, interruptions, and multi-day tasks. The reconstruction engine links fragments back into coherent workflow instances.
- Keystroke content (no keylogging)
- Screen content or recording
- Document content or text
- Personal communications
- Passwords or auth inputs
- Browser URLs or web content
- Audio or video
- Personal device activity
Alpyx captures the behavioural meta-layer of work, not the content of work. The distinction is architectural, not administrative — the agent is not capable of capturing content, and reconstruction operates via privacy-preserving aggregation, not per-individual sequences.
Technical Architecture
Lightweight at the endpoint.
Encrypted before it touches disk.
Silent enterprise installation
Deploys via Group Policy, SCCM, Intune, or manual install. No end-user action required. No workflow disruption.
Minimal endpoint impact
OS-level background process designed for sub-2% CPU and under 50MB RAM on standard enterprise hardware. Fully invisible to end users during normal operation.
Behavioural metadata only
Signals captured and structured into sequences in real time. No keylogging, no screen content recording, no document reading. Architectural, not optional.
AES-256 at source
Data encrypted on the endpoint immediately upon collection, before any local storage. The encryption key is held by the client. Alpyx does not generate, hold, or transmit keys.
Compressed time-series store
Encrypted time-series encoding optimised for event streams, achieving 40–60× compression vs. raw JSON logs. Stays within the client network perimeter at all times.
Local admin console
IT administrators manage deployment, scope, and inclusion/exclusion through a local console. revFADP documentation and works council templates ship in the package.
Tier 2 · The Intelligence Layer
Once the dataset has depth,
the analytics activate.
After 60–90 days of capture, the analytical layer unlocks bottlenecks, deviations, rework loops, and automation candidates — quantified against your own baseline. Pricing: CHF 80–150 per user per month on top of Tier 1.
Talk to us about Tier 2- Bottleneck detection
- Process deviation analysis
- Cycle time, rework, and time-waste mapping
- Cross-role and cross-team comparison
- Operational expertise mapping
- Automation-readiness scoring
- Longitudinal trend analysis (12+ months)
Deployment Roadmap
A 12-month engagement, structured in four phases.
Architecture, prototype, legal framework
Agent deployment, AES-256 store configuration, revFADP compliance review, works council documentation, and academic partner alignment (EPFL/ETHZ/ZHAW).
Pilot deployment and data collection
10–50 users per pilot site. 90+ days of capture. Real-time monitoring of agent stability across enterprise hardware. Tier 1 fully operational.
Workflow reconstruction and analytics
Boundary detection, taxonomy validation, and the Tier 2 analytics layer activate. The dataset has the depth required for cycle-time, deviation, and rework analysis.
ROI analysis and dissemination
Quantified business value at pilot sites: inefficiencies identified, time savings, automation candidates discovered. 10–25% cycle-time reduction is the benchmark target.