Bastion

AI-powered network penetration testing for macOS. Bastion combines pure-Swift network scanning with multi-backend AI orchestration to discover devices, identify vulnerabilities, chain exploits, map lateral movement paths, and generate professional security reports -- all from a native SwiftUI application on Apple Silicon.

Written by Jordan Koch (@kochj23).

---

Architecture

graph TB
    subgraph App["Bastion.app (SwiftUI)"]
        direction TB
        subgraph Views["Views"]
            Dashboard[Dashboard]
            DeviceList[Device List]
            DeviceDetail[Device Detail]
            AttackLog[Attack Log]
            AIInsights[AI Insights]
            Vulns[Vulnerabilities]
            Settings[Settings]
        end

        subgraph AI["AI Backend Manager"]
            Local["Ollama / MLX / TinyLLM\nTinyChat / OpenWebUI"]
            Cloud["OpenAI / Google / Azure\nAWS Bedrock / IBM Watson"]
        end

        subgraph Scanner["Network Scanner"]
            CIDR["CIDR Parser"]
            PortScan["TCP Port Scanner\n23 common ports"]
            ServiceFP["Service Fingerprinter\nBanner Grabbing"]
        end

        subgraph Engine["Security Engine"]
            CVEDb["CVE Database\nNIST NVD"]
            Orchestrator["AI Attack\nOrchestrator"]
            ExploitGen["AI Exploit\nGenerator"]
            Chainer["Vulnerability\nChainer"]
            LateralMap["Lateral Movement\nMapper"]
            MITRE["MITRE ATT&CK\nMapper"]
            PostComp["Post-Compromise\n10 Phases"]
            Anomaly["Anomaly Detector\nCoreML"]
            Monitor["Continuous\nMonitor"]
            Timeline["Forensic Timeline\nReconstructor"]
            Remediation["Remediation\nScript Generator"]
            PDF["PDF Report\nGenerator"]
        end

        subgraph Exploits["Exploit Modules"]
            SSH[SSH] & SMB[SMB] & DNS_M[DNS] & LDAP[LDAP] & Web[Web] & Creds[Default Creds]
        end

        Guard["Ethical AI Guardian"]
        Widget["WidgetKit Extension"]
    end

    Views --> Scanner
    Views --> AI
    AI --> Orchestrator
    AI --> ExploitGen
    Scanner --> Engine
    CIDR --> PortScan --> ServiceFP
    Engine --> Exploits
    Engine --> Guard
    Engine --> Widget

Loading

---

Features

Network Reconnaissance

Capability	Detail
Host Discovery	CIDR-based scanning (/24 and /16 subnets) with concurrent TCP connect probes
Port Scanning	23 common ports (FTP, SSH, Telnet, HTTP, HTTPS, SMB, RDP, databases)
Service Fingerprinting	Banner grabbing and version detection via raw TCP connections
OS Detection	Heuristic identification based on exposed service combinations
Quick Scan	Fast top-10-port sweep mode
DNS Resolution	Reverse DNS hostname lookup

Built on Darwin BSD sockets and the Network framework (NWConnection) -- no dependency on nmap or external tools.

AI-Powered Attack Orchestration

• AI Attack Orchestrator -- Analyzes the full threat landscape, ranks targets by exploitability, predicts success probabilities, and identifies multi-step attack chains.
• AI Exploit Generator -- Reads CVE descriptions and produces proof-of-concept exploit code (Python, Bash, Ruby) tailored to target and vulnerability. Self-improvement loop analyzes patterns every 10 runs.
• Vulnerability Chainer -- Identifies multi-step exploitation paths: info-disclosure-to-privilege-escalation, SQLi-to-RCE, path-traversal-to-credential-theft, XSS-to-admin-takeover, and exploit-to-persistence chains.
• Lateral Movement Mapper -- Discovers trust relationships (SSH key reuse, shared credentials, flat network segmentation) and builds single-hop and multi-hop pivot paths.

Post-Compromise Assessment (10 Phases)

Connects to a target over SSH and performs deep forensic inspection:

Phase	Module	What It Detects
1	Rootkit Detector	Kernel and userland rootkits
2	Suspicious User Detector	UID-0 accounts, empty passwords, anomalous shells
3	Backdoor Detector	Unauthorized listening ports and services
4	Hidden Process Detector	Processes hidden from ps / /proc
5	Binary Integrity Checker	Modified system binaries (trojaned ls, ps, etc.)
6	Persistence Detector	Cron jobs, init scripts, authorized_keys
7	Kernel Module Analyzer	Suspicious or unsigned kernel modules
8	Log Tampering Detector	Cleared logs, timestamp gaps, truncated files
9	Network Sniffer Detector	Promiscuous interfaces and packet-capture tools
10	AI Analysis	Natural-language forensic summary of all findings

CVE Database and MITRE ATT&CK

• Downloads and caches critical/high-severity CVEs from NIST NVD
• Maps services to known vulnerabilities with CVSS scores
• All findings mapped to MITRE ATT&CK techniques and tactics
• Exportable ATT&CK Navigator JSON for heatmap visualization
• Covers all 14 ATT&CK tactics from Reconnaissance through Impact

Continuous Monitoring and Anomaly Detection

• Scheduled scans at configurable intervals with baseline diffing
• CoreML-based anomaly detector learns normal device behavior and flags deviations
• macOS notifications for security alerts
• Full scan history with timeline tracking

Reporting and Remediation

• PDF reports via PDFKit: title page, executive summary, network overview, per-device vulnerability details, AI analysis
• Remediation scripts -- Hardening bash scripts per device covering SSH, web, SMB, and DNS
• Forensic timeline -- Rebuilds attacker activity sequence from post-compromise evidence with AI-generated narrative

Exploit Modules

Module	Protocol / Target
SSHModule	SSH brute force, key auth
SMBModule	SMB/CIFS, EternalBlue
DNSModule	DNS zone transfer, cache poisoning
LDAPModule	LDAP enumeration, anonymous binds
WebModule	HTTP/HTTPS, SQLi, XSS, path traversal
DefaultCredsModule	Common default credentials

Unified AI Capabilities

In addition to the core LLM backends, Bastion includes a UnifiedAICapabilities module that dynamically detects and routes to all available AI systems:

Capability Category	Backends
LLM	OpenAI GPT, Anthropic Claude, Ollama, MLX Toolkit, TinyLLM
Image Generation	ComfyUI, SwarmUI, Automatic1111, DALL-E
Voice & Audio	F5-TTS voice cloning, System TTS, cloud speech APIs
Analysis	Document analysis, vision models, structured extraction
Security	Attack orchestration, CVE analysis, hardening recommendations

Multi-model load balancing

Bastion ships with the shared multi-model LLM load balancer used across Jordan Koch's apps. Instead of pinning to a single backend, it discovers every model available on the machine and spreads work across a normalized pool, health-gating unreachable backends and failing over automatically.

Three persisted toggles (in Settings → AI Load Balancer) compose the pool:

Toggle	What it adds to the pool
Use all local models	Every discovered Ollama model + locally-installed MLX models
Enable all frontier models	OpenRouter's full frontier model list (bring your own key, stored in Keychain)
Route through Nova Gateway (optional)	Nova's OpenAI-compatible gateway (127.0.0.1:18792), which inherits Nova's own routing

The balancer supports round-robin and least-busy selection policies, tracks in-flight requests per model, and skips any backend that fails its health check. Nova is never required — the balancer works with zero Nova using local (Ollama/MLX) and/or frontier (OpenRouter) models alone; the Nova Gateway is one optional backend, and a failed health check simply drops it from the pool while everything else keeps working.

graph LR
    Finding["Security finding"] --> Prompt["explainPrompt(for:)\n(pure + redacts secrets)"]
    Prompt --> Balancer["LLM Load Balancer"]

    subgraph Toggles["Persisted toggles"]
        T1["Use all local models"]
        T2["Enable all frontier models"]
        T3["Route through Nova Gateway (optional)"]
    end
    Toggles --> Pool

    subgraph Pool["Discovered model pool"]
        Ollama["Ollama models"]
        MLX["MLX models"]
        OpenRouter["OpenRouter frontier"]
        Nova["Nova Gateway (optional)"]
    end

    Balancer -->|health-gate + round-robin / least-busy| Pool
    Pool -->|"first healthy, fall through on failure"| Result["Plain-English summary"]
    Result -.->|no backend reachable| Disabled["Feature disabled with a clear reason\n(core security functions unaffected)"]

Loading

Explain this finding (AI)

Every security finding gets an Explain action that routes the finding through the balanced LLM and returns a plain-English summary: what it means, why it matters, and suggested remediation, shown in a sheet.

• The prompt is built by a pure, network-free FindingExplainer.explainPrompt(for:) so it is fully unit-tested.
• Secrets are never sent off-box. The prompt builder redacts obvious secret-like content (passwords, API keys, tokens, bearer/basic auth, HTTP basic-auth in URLs, AWS access-key ids, PEM private-key blocks) before anything reaches the LLM.
• The feature degrades gracefully: when no LLM backend is reachable it is disabled with a clear reason and never crashes, and it never blocks Bastion's core security functions on LLM availability.

WidgetKit Extension

Small, medium, and large widgets displaying security score, vulnerability breakdown, devices at risk, last scan time. Auto-syncs via App Group group.com.jkoch.bastion.

Ethical AI Safeguards

Content monitoring with 100+ prohibited-use patterns, automatic blocking, crisis resource referrals, hashed audit logging. Terms of Service enforced at every launch.

---

AI Backends

Backend	Type	Default Endpoint	Notes
Ollama	Local	localhost:11434	Preferred default; pull any GGUF model
MLX	Local	Python subprocess	Apple Silicon native via mlx-lm
TinyLLM	Local	localhost:8000	OpenAI-compatible server
TinyChat	Local	localhost:8000	Fast chatbot with streaming
OpenWebUI	Local	localhost:8080 or :3000	Self-hosted AI platform
OpenAI	Cloud	OpenAI API	GPT-4o
Google	Cloud	Vertex AI	Vision, Speech
Azure	Cloud	Cognitive Services	Full Azure AI suite
AWS	Cloud	Bedrock, Rekognition, Polly	Full AWS AI suite
IBM Watson	Cloud	NLU, Speech, Discovery	Natural language understanding

Auto mode probes each backend in priority order (Ollama first) and selects the first available. API keys stored in macOS Keychain.

---

Responsible Use

Bastion is designed exclusively for authorized security testing, penetration testing engagements, CTF competitions, and educational purposes. Scanning is restricted to RFC 1918 local IP addresses (192.168.x.x, 10.x.x.x, 172.16-31.x.x). A legal warning dialog with explicit acknowledgment is required at every application launch.

Unauthorized access to computer systems is illegal under the Computer Fraud and Abuse Act (CFAA), the Computer Misuse Act, and equivalent legislation in your jurisdiction. Always obtain proper written authorization before testing.

---

Installation

Bastion is distributed as a DMG installer. It is not available on the Mac App Store.

From DMG (recommended for most users)

1. Download the latest .dmg from Releases.
2. Open it and drag Bastion into your Applications folder.
3. Launch it from Applications. That's it — no Xcode, no toolchains, nothing else to install.

See "Bastion can't be opened because the developer cannot be verified"? That means you have a build that isn't yet Developer-ID-signed and notarized. To open it anyway:

• macOS 14 and earlier: Control-click (right-click) the app → Open → Open.
• macOS 15 (Sequoia) / 26 and later: double-click it, dismiss the dialog, then open System Settings → Privacy & Security, scroll down, and click Open Anyway.
• Or from Terminal: xattr -dr com.apple.quarantine "/Applications/Bastion.app"

Notarized releases open with no prompt at all — maintainers, see RELEASE.md.

From Source

Requires Xcode 16 or later and macOS 13.0 Ventura or later. Because the app bundles MLX for on-device LLM inference, the build compiles Metal GPU shaders, which needs Apple's Metal Toolchain — a component Xcode 16 no longer ships by default. Install it once:

xcodebuild -downloadComponent MetalToolchain
# (or in Xcode: Settings → Components → Metal Toolchain → Get)

Then build:

git clone git@github.com:kochj23/Bastion.git
cd Bastion
xcodebuild -project Bastion.xcodeproj -scheme Bastion -configuration Release build

Skipping the Metal Toolchain step produces a wall of CompileMetalFile … cannot execute tool 'metal' due to missing Metal Toolchain errors from the mlx-swift dependency. That's the missing component, not a problem with the project.

App sandbox is disabled for network scanning, SSH connections, and raw socket access.

AI Backend Setup (Optional)

# Ollama (recommended for local AI)
brew install ollama && ollama serve && ollama pull mistral:latest

# TinyLLM (lightweight OpenAI-compatible server)
pip install tinyllm && tinyllm serve

---

Keyboard Shortcuts

Shortcut	Action
Cmd+N	New Scan
Cmd+S	Stop Scan
Cmd+Q	Quick Scan
Cmd+R	Run AI Attack Plan
Cmd+Option+Shift+X	Full Assault Mode
Cmd+.	Emergency Stop
Cmd+Option+B	AI Backend Settings
Cmd+1 through Cmd+5	Switch view tabs

---

Technical Details

• Language: Swift 5.9, SwiftUI
• Minimum OS: macOS 13.0 (Ventura)
• Architecture: Apple Silicon native (arm64), Universal binary supported
• Sandbox: Disabled (com.apple.security.app-sandbox = false) -- network scanning, SSH connections, and raw socket access require full system permissions
• App Group: group.com.jkoch.bastion (shared data with WidgetKit extension)
• Network layer: Darwin BSD sockets via the Network framework (NWConnection); no external tool dependencies
• AI integration: OpenAI-compatible HTTP APIs for all local and cloud backends; Python Process subprocess for MLX
• CVE data: Cached in ~/Library/Application Support/Bastion/CVE/
• PDF generation: Native PDFKit + CGContext rendering
• Anomaly detection: CreateML / CoreML for on-device behavior profiling
• Concurrency: Swift structured concurrency (async/await, TaskGroup) throughout

---

Project Structure

Bastion/
  BastionApp.swift              App entry point, legal warning, main UI
  AI/
    AIBackendManager.swift       Multi-backend AI manager (10 backends)
    AIAttackOrchestrator.swift   Network-wide attack planning
    AIExploitGenerator.swift     CVE-to-exploit code generation
  AICapabilities/
    UnifiedAICapabilities.swift  Unified capability detection and routing
    ImageGenerationUnified.swift Image generation backends
    VoiceUnified.swift           Voice/speech backends
    AnalysisUnified.swift        Analysis backends
    SecurityUnified.swift        Security-specific AI orchestration
  Models/
    Device.swift                 Device, Port, Service, Vulnerability models
    CVE.swift                    CVE data model
    AttackResult.swift           Attack outcome tracking
    CompromiseReport.swift       Post-compromise assessment report
  Security/
    NetworkScanner.swift         Pure-Swift CIDR scanner
    CVEDatabase.swift            NIST NVD downloader and cache
    ServiceFingerprinter.swift   Banner grabbing and version detection
    ComprehensiveDeviceTester.swift  Full device security audit
    VulnerabilityChainer.swift   Multi-step exploit chain builder
    LateralMovementMapper.swift  Network pivot path discovery
    MITREATTACKMapper.swift      ATT&CK technique/tactic mapping
    ContinuousMonitor.swift      Scheduled scan and alert engine
    AnomalyDetector.swift        ML-based behavior anomaly detection
    TimelineReconstructor.swift  Forensic attack timeline builder
    RemediationScriptGenerator.swift  Hardening script output
    ExploitModules/
      SSHModule.swift            SSH-specific testing
      SMBModule.swift            SMB/CIFS testing
      DNSModule.swift            DNS testing
      LDAPModule.swift           LDAP testing
      WebModule.swift            HTTP/HTTPS testing
      DefaultCredsModule.swift   Default credential testing
    PostCompromise/
      PostCompromiseModule.swift  10-phase assessment orchestrator
      RootkitDetector.swift      Rootkit scanning
      BackdoorDetector.swift     Backdoor scanning
      HiddenProcessDetector.swift  Hidden process detection
      SuspiciousUserDetector.swift  User account analysis
      PersistenceDetector.swift  Persistence mechanism scanning
      KernelModuleAnalyzer.swift Kernel module inspection
      LogTamperingDetector.swift Log integrity checking
      NetworkSnifferDetector.swift  Promiscuous mode detection
      BinaryIntegrityChecker.swift  System binary hash verification
      BinaryHashDatabase.swift   Known-good hash database
  Utilities/
    SSHConnection.swift          SSH client for post-compromise
    PDFGenerator.swift           Enterprise PDF report output
    WidgetDataSync.swift         Widget data sync via App Group
    SafetyValidator.swift        RFC 1918 enforcement and rate limiting
    ModernDesign.swift           Glassmorphic UI styling
  Views/
    DashboardView.swift          Main dashboard with scan controls
    DeviceListView.swift         Discovered device list
    DeviceDetailView.swift       Per-device detail inspector
    AttackLogView.swift          Attack execution log
    AIInsightsView.swift         AI recommendations and plans
    VulnerabilitiesView.swift    Vulnerability browser
    SettingsView.swift           Backend and scan configuration

---

How Bastion Compares

Feature	Bastion	Metasploit	Burp Suite
AI-powered exploit selection	Yes (10 backends)	No	No
AI exploit code generation	Yes	No	No
Vulnerability chain analysis	Yes	No	No
Lateral movement mapping	Yes	Limited	No
MITRE ATT&CK mapping	Yes	Limited	No
Post-compromise forensics	Yes (10-phase)	Post modules	No
Native macOS app	Yes (SwiftUI)	No (CLI/Java)	No (Java)
Local AI (no cloud required)	Yes	N/A	N/A
Apple Silicon native	Yes	No	No
WidgetKit dashboard	Yes	No	No
PDF report generation	Yes	Yes	Yes
Free and open source	Yes (MIT)	Community Edition	No

---

Testing

200 tests across 11 test classes:

Test Class	Tests	Coverage
ComprehensiveTests	60	Cross-module integration, end-to-end workflows, all model types
DeviceModelTests	24	Device, OpenPort, ServiceInfo, risk levels, security score
CompromiseReportTests	22	Post-compromise findings, assessment logic, all finding types
MITREATTACKMapperTests	16	ATT&CK technique/tactic mapping, Navigator JSON export
AIBackendTests	15	Backend enum, error types, AttackPlan, PostExploitationPlan
CVEModelTests	15	CVE severity mapping, Codable round-trips, VulnerabilitySeverity
SafetyValidatorTests	14	RFC 1918 enforcement, public IP blocking, rate limiting
AttackResultTests	10	AttackResult, AttackStatus, AttackType, ScanResults
VulnerabilityChainerTests	8	Chain probability calculation, chain types
LateralMovementTests	8	Trust relationships, movement paths, multi-hop filtering
NetworkErrorTests	8	NetworkError, CVEDatabaseError, CVEMetadata

The ComprehensiveTests suite (added May 2026) covers 60 tests across five categories: unit, security, integration, functional, and frame/enum tests. It validates Codable round-trips for Device and CVE, compromise assessment logic at all four confidence levels, vulnerability chain probability calculation, MITRE tactic ordering, and every finding type in the post-compromise module.

xcodebuild test -project Bastion.xcodeproj -scheme Bastion \
  -configuration Debug -destination 'platform=macOS' \
  -only-testing BastionTests

---

Download

Download the latest release: Bastion on GitHub Releases

---

More Apps by Jordan Koch

App	Description
NMAPScanner	Network security scanner with AI threat detection
MLXCode	Local AI coding assistant for Apple Silicon
URL-Analysis	Network traffic analysis and URL monitoring
TopGUI	macOS system monitor with real-time metrics
rtsp-rotator	RTSP camera stream rotation and monitoring

View all projects

---

License

MIT License -- see LICENSE.

Ethical usage required -- see ETHICAL_AI_TERMS_OF_SERVICE.md.

Copyright (c) 2026 Jordan Koch. All rights reserved.

---

Written by Jordan Koch (@kochj23).

Disclaimer: This is a personal project created on my own time. It is not affiliated with, endorsed by, or representative of my employer.