Security
SSH keys encrypted via AES-GCM (Android) and Keychain (iOS). Never transmitted.
PIN hashed with PBKDF2-HMAC-SHA256, 100,000 iterations. Built-in brute force protection.
Zero data sent to our servers. No telemetry, no analytics.
Screen protection enabled. Anti-capture and anti-recording by default.
Data Protection
Sensitive data storage
All sensitive data is stored via Flutter Secure Storage: AES-GCM encryption on Android (hardware Android Keystore) and Keychain on iOS. The master encryption key is stored in the device's hardware keystore, making extraction impossible even with root access. Protected data: SSH private keys, PIN hash, SSH server fingerprints, audit log, Wake-on-LAN configurations, and command history. No secrets are hardcoded in the source code — a full codebase scan confirms this.
Authentication & Connections
SSH connection and TOFU verification
Standard SSH2 protocol. Preferred key algorithm: Ed25519 (256 bits). On first connection, the server's SHA-256 fingerprint is displayed. The user must manually confirm it. The fingerprint is then stored and automatically verified. If it changes: red alert with an explanation of the interception risk (MITM). Fingerprint comparison is performed in constant time (bitwise XOR) to prevent timing attacks. 16 weak SSH algorithms are blocked at the software level (SHA-1, CBC, arcfour, 3DES and their variants). Only modern, recommended algorithms are accepted. Private keys are loaded into a dedicated SecureBuffer. After use, memory is zeroed out. The SSH worker re-reads the key from secure storage on each new connection — no persistence in memory between sessions. Cryptographic operations run in a separate isolate (background thread).
Local authentication and brute force protection
PIN code: minimum 8 digits (100 million combinations). Hashed with PBKDF2-HMAC-SHA256 over 100,000 iterations, combined with a unique random salt of 32 bytes. The plaintext PIN is never stored. Hash comparison is performed in constant time. Brute force protection: lockout after 5 failed attempts, with exponential backoff (30s, 60s, 120s, capped at 300s). The counter is monotonic: it only resets after a correct PIN, never by delay expiration. Biometrics: native system API, biometric data never leaves the device. Strict mode: biometrics only, no fallback to the device system PIN. Automatically invalidated when the app goes to background. Configurable auto-lock (5, 10, 15 or 30 minutes). Loading screen at startup until the PIN state is fully loaded, preventing any temporary bypass.
Data leak protection
Command history: regex filtering excludes AWS keys, JWT/Bearer tokens, API keys, command-line passwords, and sensitive variables. Automatic expiration after 90 days. Manual wipe available. Clipboard: automatically cleared 30 seconds after any sensitive data copy. Also cleared when the app goes to background via the native Android API. Production logs: all debug calls gated by kDebugMode. Zero logs in release builds. No hostnames, IPs, or identifiers in production. Go logs (Tailscale) are filtered to prevent OAuth token or authentication URL leaks.
Screen protection
Android: FLAG_SECURE enabled by default. Blocks screenshots and screen recording. The app appears as a black screen in the recent apps selector. Can be disabled in settings. iOS: automatic masking screen when the app goes to background. Prevents content capture in the app switcher. Can be disabled in settings.
Compromised device detection
At startup, the app checks for the presence of root (Android) or jailbreak (iOS) characteristic paths and files. If detected: a warning banner explains the risk to SSH key security. This is an informational measure, not a blocking one — the user can continue at their own discretion.
Anti-tampering protection
Integrated runtime protection detects 12 types of threats: root/jailbreak, attached debugger, code injection hooks, emulator, APK tampering, unofficial store installation, missing obfuscation, no device screen lock, developer mode enabled, and ADB connection. Two configurable modes: "Warn" (event logged in the encrypted audit journal) or "Block" (alert screen preventing app usage). Automatically disabled in debug mode to avoid false positives during development. Configurable by the user in Settings > Security.
Features & Integrations
Audit log
Automatically recorded events: SSH connection (success/failure), disconnection, reconnection, authentication failure, SSH key import/deletion, PIN creation/deletion, server fingerprint change. Each entry is chained with a SHA-256 hash of the previous entry. Tampering with any single entry invalidates all subsequent entries — integrity is verifiable at any time. Stored encrypted in secure storage, compact JSON format with timestamps. FIFO rotation at 500 entries. Never leaves the device.
SFTP transfers and key import
SFTP transfers limited to 30 MB per file. Chunk-based streaming, no full loading into memory — prevents saturation attacks (DoS/OOM). Directory traversal detection. SSH key import: format validated, size limited to 16 KB (a normal key is under 5 KB). Abnormally large files are blocked. Immediate storage in secure storage after import.
Tailscale integration
WireGuard VPN connection via libtailscale (native Go library, Kotlin channel). OAuth authentication URLs are never logged in plaintext — only their length is recorded in debug mode. Peer public keys are truncated in logs. Error messages are generic. URL validation: HTTPS scheme only before opening in the browser. No Tailscale token stored on the Flutter side — all token-related code has been removed.
Supply chain security
6 security-critical packages are locked to exact versions (no automatic updates via the ^ operator): SSH library, cryptographic primitives, secure storage, biometric authentication, and anti-tampering framework. APK signing is mandatory for release builds — the build fails without a production keystore. Code obfuscation is enabled on every release build: class names, methods, and variables are made unreadable, and debug symbols are separated and never included in the distributed binary.
Permissions and architecture
Android: ADB backup disabled (allowBackup=false), services not exported, VPN service protected by system permissions. Only necessary permissions requested: network (SSH/WoL), biometric sensor (optional), local storage (encrypted credentials). SSH cryptographic operations in a separate isolate (background thread). Request identifiers use cryptographically random UUID v4. Interface translated into 5 languages, no sensitive strings hardcoded.
Transparency
Known limitations
The following limitations are documented for transparency. Honest disclosure of trade-offs is preferable to concealment. The SSH private key passes through memory as a Dart String (immutable), potentially persisting until garbage collection. Low impact: requires a rooted device with memory access. Mitigated by re-reading from secure storage on each connection. The Dart garbage collector may create temporary copies in memory. Same prerequisites. Root detection can be bypassed by advanced tools (Magisk Hide). Informational measure only. The generated Ed25519 key uses cipher=none. Acceptable as long as it remains in secure storage (AES/Keychain encrypted). The PIN hash is stored in SharedPreferences (locally accessible). Mitigated by PBKDF2 making offline brute force impractical.
Audit results
All audits are internal reviews conducted with specialized security analysis tools. No external audit has been performed to date. White-box audit V1.5.1 (February 2026): score improved from 6.5 to 8.5/10, 9 fixes applied. STRIDE analysis (February 2026): 22 threats identified, 8 risks validated, 12 mitigations implemented at 100%. Quality audit (February 2026): 4 critical bugs fixed, dead code removed, 92/92 tests pass. Advanced security audit (February 2026): 62 findings, 0 remotely exploitable vulnerability. 27 fixed, 11 mitigated/documented. Red Team + defensive hardening (February 2026): 25 attack vectors identified and analyzed, 26 security modules created, 30 existing files hardened. 738 automated tests — 0 regressions.
Related
All system commands go through a single, auditable execution point.
PIN hashed with PBKDF2-HMAC-SHA256, 100,000 iterations. Built-in brute force protection.
Zero data sent to our servers. No telemetry, no analytics.
Command injection protection validated by audit.
Secure Architecture
Secure architecture
All system commands go through a single centralized class, creating an audit and control point for every execution. Arguments are passed as a separated list, preventing shell metacharacter interpretation. Default 120-second timeout on all commands. Exception handling: timeouts and process errors are caught cleanly.
Fail-closed circuit breaker
If 3 consecutive errors occur on a critical service, it is automatically shut down rather than allowed to continue in a degraded state. This follows the security-by-default-denial principle: when in doubt, block rather than pass.
Startup integrity checks
At every launch, Chill runs integrity checks to detect the presence of code injection tools or debuggers attached to the process. If a threat is detected, the app shuts down immediately (fail closed). These protections are automatically disabled in debug mode to avoid false positives during development.
Privilege escalation
Single mechanism adapted to each operating system: Windows: temporary PowerShell script executed via the -File parameter (prevents PowerShell from interpreting the script content). Linux: temporary bash script executed via pkexec (polkit). Arguments as separated list. macOS: temporary script executed via osascript with POSIX escaping (single-quote encapsulation). In all cases: unique temporary directories, 700 permissions (owner only), systematic deletion after execution. Only one password prompt per operation — all admin commands are grouped into a single script.
Authentication & Protection
Local authentication and brute force protection
PIN code: minimum 8 digits. Strict regex validation in business logic. Hashed with PBKDF2-HMAC-SHA256 over 100,000 iterations + random 16-byte salt (cryptographically secure generator). Constant-time comparison (bitwise XOR). Brute force protection: lockout after 5 attempts, exponential backoff (30s, 60s, 120s, 300s). Monotonic counter — reset only by correct PIN. Counters are persisted and survive restarts. Loading screen at startup prevents any bypass. Automatic transparent migration from legacy hash formats.
Automated Configuration
Automated SSH configuration
Temporary scripts in unique directories with 700 permissions, deleted after execution. Semantic exit codes for precise error diagnosis. Post-installation verification of the SSH service. Firewall rule created only if none exists. Supported systems: Windows (OpenSSH Capabilities), Linux (automatic apt/dnf/pacman detection, systemd), macOS (systemsetup).
Wake-on-LAN configuration
Same secure model as SSH. Strict network interface name validation via regex (alphanumeric, hyphens and dots, max 15 characters). Windows adapter names escaped for PowerShell. Systemd service created on Linux for persistence. All admin commands grouped in a single pkexec call.
Tailscale integration
Communication with the Go daemon (chill-tailscale) via JSON over stdin/stdout. The daemon binary's SHA-256 hash is verified at startup — if the hash doesn't match, the daemon is not launched (fail closed). Every IPC message is authenticated with HMAC-SHA256 and includes a timestamp to reject replayed messages. URL validation (HTTPS only) before opening in the browser. Guard against concurrent calls during daemon restarts. Clean shutdown with 3-second timeout and forced kill as fallback. Defensive JSON parsing: try-catch on every response, invalid peers filtered. Debug logs only, inactive in production.
OS Security & Hardening
OS security toggles
Built-in interface for hardening your operating system without using a terminal. One-click toggles to enable security protections: 8 toggles on Windows (firewall, disable RDP, disable SMBv1, anti-ransomware, BitLocker...), 7 on Linux (UFW firewall, Fail2Ban, secure network settings, disable root login...), 8 on macOS (application firewall, stealth mode, FileVault, Gatekeeper...). Disabling a critical protection requires progressive confirmation: a reflection delay before the button activates, plus manual entry of a confirmation word. This prevents accidental one-click deactivation of important protections.
System security checkup
Read-only 12-point scan analyzing your PC's security posture with a score. Checks include: firewall active, pending updates, disk encryption, antivirus signatures, quick malware scan, startup programs, suspicious scheduled tasks, active network connections, failed login attempts, user accounts, disk space, and OS-specific settings. Result: score out of 12 with personalized recommendations. Everything is 100% local — no data is sent over the network.
Sandbox and minimal permissions
Each platform uses its native sandboxing mechanism to restrict Chill's access to only what is strictly necessary. Linux: AppArmor profile restricting file, network, and system capability access. macOS: minimal entitlements with Hardened Runtime enabled, preventing dynamic code injection. Windows: application control policy verifying component signatures. Only explicitly authorized directories and sockets are accessible.
SSH key rotation
SSH keys generated by Chill have a configurable lifespan (30 days by default). As expiration approaches, the user is notified. Manual rotation is available at any time from the interface.
Injection & Leak Protection
Command injection protection
All injection vectors identified by the audit have been fixed: Windows elevation: -File parameter instead of -Command, preventing script content interpretation. macOS elevation: dedicated POSIX escaping function (single-quote encapsulation). Network interface names: centralized strict regex validation across all functions. Windows adapter names: apostrophe escaping for PowerShell. WoL PowerShell commands: single quotes for interpolated values.
Information leak protection
Generic error messages shown to the user, technical details sent to debug logs only. Clipboard automatically cleared 30 seconds after copy. All silent catch blocks replaced with informative logs (6 blocks fixed). Zero sensitive information in production logs.
Transparency
Known limitations
The following limitations are documented for transparency. The entered PIN passes through memory as a Dart String (immutable), persisting a few milliseconds until garbage collection. Very low impact. The PIN hash and salt are stored in SharedPreferences (unencrypted). Mitigated by PBKDF2 making offline brute force impractical. The Process.run timeout does not guarantee child process termination (Dart limitation). Documented, future improvement planned. TOCTOU on temporary scripts: a local process could theoretically modify the script between creation and execution. Mitigated by 700 permissions. The Tailscale daemon binary may be resolved via the system PATH as fallback. A warning is logged.
Audit results
All audits are internal reviews. No external audit has been performed to date. Architectural context audit (February 2026): complete mapping of 12 modules, 7 actors, 13 trust boundaries, 6 end-to-end workflows. Full security audit (February 2026): 38 findings (4 critical, 8 high, 14 medium, 12 low). 27 fixed, 11 mitigated/documented. 61/61 tests. Quality audit (February 2026): 34 findings (6 critical, 14 important, 14 minor). All fixed. 42/42 tests. Red Team + defensive hardening (February 2026): 58 attack vectors identified and analyzed, 44 security modules created, 3 existing files hardened. 869 automated tests — 0 regressions.
Related
HUYNH Kevin
Creator of Chill