SYSTEM_STATUS: NATIVE

The machine
inside the machine.

OmniCap isn't a cloud service in a desktop wrapper. It is a strictly air-gapped cognitive engine built directly onto Apple Silicon MLX.

The End of the Cloud API

See exactly how your data moves.

Standard Agent
Latency: ~800ms
OmniCap Native
Latency: < 12ms

The 3-Layer Anatomy

Hover to isolate the internal pipeline.

01. Sensory Hook

Instantly reads accessibility frames via `AXUIElement` and `CGEvent` without polling, achieving a sub-1% CPU idle footprint.

02. MLX Compute

Executes optimized INT4/FP4 local models directly on the Neural Engine via Swift C++ bindings. Bypasses Python completely.

03. Temporal Graph

Appends relational memories to an encrypted local SQLite database, creating a deeply connected timeline of your workflows.

Inside a Memory Node

Standard AI uses rigid vector databases. OmniCap uses a dynamic Temporal Graph. Hover the node to see exactly what metadata is captured locally with every frame.

NODE: 9X4-AF212:43:02 PM
AppXcode
WindowNativeMLXEngine.swift
Vectors1,024 embeddings generated

Technical FAQ

The three most common questions from engineers.

OmniCap uses aggressive delta-compression and vector deduplication. 30 continuous days of 4K screen history averages around 150MB of local storage. You can set auto-pruning rules directly in the preferences.

No. The AppKit CGEvent hooks idle at < 0.1% CPU because they do not rely on continuous polling. The MLX inference engine only wakes up when specifically triggered by the Accessibility tree delta.

Absolutely not. OmniCap is 100% air-gapped. We do not use Mixpanel, Crashlytics, or any cloud telemetry SDKs. You can verify this by monitoring our binary with Little Snitch—it won't even try to phone home.

Verify it yourself.

Audit the SQLite database on your own machine. Download the purely air-gapped V1 release.