Skip to Content
Future Improvements

Future Improvements & Roadmap

The development roadmap for Cerberus focuses on intelligent automation, machine learning assistance, and expanding endpoint coverage.


1. Smart Adaptive Scanning Speed

The Concept

Currently, scan concurrency parameters (such as RustScan batching and Nmap min-parallelism) use tuned default values (--min-parallelism 64, --max-rtt-timeout 500ms).

In future releases, Cerberus will implement Dynamic Throughput Tuning:

  • Automatic RTT Calibration: The scanner probes the local default gateway with 5 test ping datagrams to measure baseline latency and jitter.
  • Backpressure Detection: If packet loss exceeds 2% or round-trip time spikes by more than 50%, the engine automatically scales back packets per second (pkts/s).
  • Device-Aware Throttling: Detects low-power embedded devices (e.g. smart bulbs, microcontrollers) and reduces port probe speed to avoid crashing sensitive embedded network stacks.
Target Adaptive Scan Efficiency
99.4%(Zero router drops)

Automatically scales from 500 pkts/s on low-end routers up to 10,000 pkts/s on enterprise switches.


2. Lightweight AI / ML Helper Model for OS & Device Guessing

The Concept

While Nmap’s TCP/IP stack fingerprinting (--osscan-guess) is powerful, it frequently struggles with modern embedded IoT appliances, streaming sticks, and custom smart home hubs where signatures are ambiguous or listening ports are sparse.

Implementation Plan

  • Embedded ONNX / TensorRT Model: Embed a compact machine learning model (< 15 MB) into the Go binary.
  • Multi-Feature Inference: Input features will include:
    • IEEE OUI vendor prefix
    • TCP initial window size, TTL, and DF flags
    • mDNS TXT record strings (e.g. model=J42d, deviceid=...)
    • Open port combination fingerprints (e.g., ports 80 + 554 + 8000)
  • High-Fidelity Classification: Produces a ranked prediction with confidence scores (e.g. "Philips Hue Bridge v2 (94% confidence)" instead of generic "Linux 3.X").

3. Passive DHCP Snooping & Raw Packet Capture

The Concept

Currently, new devices are detected passively through mDNS/SSDP announcements or operating system neighbor cache polling.

  • The Future Feature: Implementing a raw capture listener on the local interface to monitor DHCP DISCOVER and REQUEST broadcasts (UDP 67/68).
  • Instant Discovery: Because every device must negotiate DHCP to join the network, Cerberus will detect new devices the exact millisecond they connect, before they even finish obtaining an IP address.

4. Remote LAN Agent Ingestion & Enterprise Fleets

In the current release, the eBPF agent ingest endpoint is intentionally restricted to loopback (127.0.0.1:8080) to ensure absolute security out of the box.

  • Remote Ingest (Tracked in Issue #8): Adding mutual TLS (mTLS) and token rotation to allow remote Linux servers across the office or home lab to stream eBPF telemetry into a central Sentinel desktop app.
  • UI Incident Console (Tracked in Issue #12): A dedicated security incident view within the Wails desktop frontend showing real-time MITRE ATT&CK process trees and kill-chain alerts.
Last updated on