PRIVATE ENGINEERING ENGAGEMENT · SECURE DEVELOPMENT · OCTOBER 2026
PentestGPT is a cybersecurity tool leveraging Artificial Intelligence to assist ethical hackers during penetration testing, but out of the box it operates strictly via raw terminal commands. At Indomitus One, we engineered its visual mission control: a local, hardened web dashboard that enables operators to control PentestGPT with any compatible LLM, governing targets, execution stages, and findings in real time from a single screen.
02 — THE TECHNICAL CHALLENGE
While PentestGPT accelerates security testing through AI reasoning, managing it exclusively through raw terminal commands creates four key operational hurdles that our dashboard resolves:
Tracking AI responses and tool outputs across multiple command-line windows makes it difficult to quickly visualize the active audit stage.
A visual dashboard featuring a 5-stage progress tree, color-categorized console filtering, and a structured findings registry.
During an engagement, it is crucial to ensure that no command or AI reasoning probe inadvertently targets IP addresses outside the authorized subnet.
Automated CIDR whitelist enforcement: if the target IP falls outside the allowed subnet range, the system blocks execution immediately.
Engagements vary: some require 100% offline local inference for strict confidentiality, while others demand switching providers on the fly.
Integrated model selector compatible with any PentestGPT model, including private offline execution without sending data to external APIs.
If an anomaly occurs during testing in a sensitive production lab, the operator needs to halt all activity instantly without orphaned processes.
Prominent emergency stop button (Kill Switch) that aborts the active mission, plus a Lock button to seal access with one click.
03 — THE DEVELOPED SOLUTION
Explore real screenshots of the platform. You can switch between the operations dashboard and the unlock screen, or click on the five numbered hotspots to inspect each module.
Where the operator enters the target IP (10.10.10.15), assigns the mission name, specifies instructions, chooses the LLM to use with PentestGPT, and triggers START or immediate STOP / KILL.
Displays at a glance the PentestGPT engine state, connection health to the configured AI provider, and real hardware telemetry: CPU (8.7%) and RAM (12.2 GB).
Divides the engagement into 5 clear stages (Initialization, Reconnaissance, Enumeration, Analysis, and Findings) to clearly see completed steps and active tasks.
Streams all incoming events in real time ([SYSTEM], [AI], [TOOL], [FINDING]) with instant category filtering without drowning in thousands of CLI lines.
Continuously displays loopback isolation (BIND: 127.0.0.1:8787), authorized subnets (SCOPE CIDR), session status, and the instant Lock button to seal the panel.
Upon startup, the access key is held strictly in RAM memory, leaving zero credential artifacts or tokens stored on disk.
Until the operator inputs the valid startup token, the UI remains blurred and all real-time WebSocket channels reject connections.
Clicking the Lock button in the top-right corner immediately invalidates the session and disconnects active telemetry streams.
04 — ARCHITECTURE & SECURITY
Built using Python (FastAPI), WebSockets, and clean web standards. Click on each layer to inspect its defensive invariants:
Standard HTML5 / CSS3 / Vanilla JS · 0 CDNs
The dashboard is served directly from the local server without loading third-party libraries, fonts, or scripts from the Internet.
Exclusive loopback binding on 127.0.0.1:8787
The software only listens on the operator’s own computer, preventing other devices on the LAN or Wi-Fi from reaching the panel.
RAM Startup Token + CSRF-Protected Sessions
Unlocking the panel requires an ephemeral key generated in RAM upon startup, leaving zero passwords or secrets persisted on disk.
Subnet whitelists with Fail-Closed policy
A defensive guardrail validating every target IP address against allowed subnets before allowing any probe to launch.
Event streaming with automated session teardown
Instant communication channel pushing task tree updates, console logs, and findings without page reloads.
Compatible with any AI model + Emergency Stop
Flexible architecture allowing operators to use any LLM compatible with PentestGPT (offline local models or remote providers) with emergency abort.
05 — METHODOLOGY & VALIDATION
Development was completed in four disciplined phases: from visual workflow mapping to three security hardening rounds and automated test suites.
We translated PentestGPT’s terminal workflow into a structured graphical interface, dividing the screen into mission control, task tree, filtered console, and findings.
We engineered the tool following minimal exposure principles so the dashboard is only accessible to the operator and strictly acts on authorized targets.
We subjected the platform to three exhaustive security reviews and hardening cycles to lock down access control and real-time streaming channels.
We validated every defensive invariant through automated test suites to ensure the platform handles invalid inputs and emergency stops flawlessly.
06 — INTERACTIVE DEMONSTRATION
Select one of the three sample missions, filter live console streams, or test the STOP button (Kill Switch) to see the dashboard’s interactive behavior.
10.10.10.15✓ AUTHORIZED CIDRLab-Recon-01LLM Model (PentestGPT)What this order executes: Identify open ports and exposed services on the authorized IP (10.10.10.15) without running active exploit payloads.
At any point, the operator can click STOP to abort the active mission immediately and cleanly terminate background processes.
We design custom management consoles and internal applications that transform complex technical processes into visual, hardened, and accessible tools:
Contact us to request a demonstration of the project or to discuss the custom software solutions your business requires.