Source-available · Rust · Safe by default
Autonomous security agent for Linux. 38 eBPF kernel hooks, 48 detectors, 23 cross-layer correlation rules. Detects attacks from firmware to userspace. Responds automatically. Rust.
How it works
Five stages. Zero overhead.
Why you need this
Attacks don't wait. Your defense shouldn't either.
Threats are constant
SSH brute-force, port scans, web scanners, credential stuffing. Every public server gets hit. Most operators discover this only after something goes wrong.
Detect, score, block. Automatically.
Inner Warden watches your logs in real time. Every threat gets an AI-scored confidence level. Above your threshold? Firewall rule added. Telegram alert sent.
You stay in control
Dry-run by default. Full audit trail. Every decision logged. Approve or deny actions from Telegram. Go live only when you trust the decisions.
See it in action
It's 2 AM. Someone brute-forces your SSH. You're asleep.
Here's what Inner Warden does while you sleep.
1
Threat contained. 4 seconds.
8 failed SSH logins in 300 seconds from 203.0.113.42. Inner Warden catches the pattern, AI confirms brute-force with 0.90 confidence. Firewall deny rule added via ufw. Your phone buzzes with the Telegram alert.
2
Honeypot traps the attacker.
A different IP tries the same trick. This time they hit the honeypot. Fake SSH session, real capture. Every credential and command logged. Session ends, IP blocked, transcript saved.
Blocked IPs are reported to AbuseIPDB and pushed to Cloudflare WAF. Your defense protects others too.
Beyond log analysis
Other tools detect and alert. Inner Warden blocks inside the kernel.
22 eBPF programs run inside the Linux kernel. They see every syscall, every process, every connection. When an attacker builds a reverse shell, the kernel itself denies the final execve. No userspace delay. No evasion possible.
🧬
38 kernel hooks
execve, connect, ptrace, setuid, bind, mount, memfd_create, mprotect, clone, and 13 more. Every security-relevant syscall monitored.
⛓️
Kill chain detection
7 attack patterns tracked per-PID: reverse shell, bind shell, code injection, shellcode execution, and 3 zero-day exploit patterns. Blocked at execve. The attacker never gets a shell.
🚫
LSM enforcement
Linux Security Modules hook, the same framework as SELinux and AppArmor. Blocks execution from /tmp and /dev/shm. Policy-gated, fail-open by default.
⚡
XDP wire-speed blocking
Malicious IPs are dropped at the network driver level before reaching the kernel stack. 10 million+ packets per second. Zero CPU overhead.
🔇
Kernel-level noise filters
137 safe processes allowlisted in BPF maps. Package managers, compilers, system daemons. Filtered before events reach userspace. Zero false positives from known-good software.
🎯
Zero-day detection
No signatures. No CVE database. Attack patterns are detected by syscall sequence: mprotect(RWX) + dup2(stdin) + execve = blocked. Works against exploits that don't exist yet.
How it protects you
Start safe. Scale up when ready.
Every feature is off by default. Enable what you need, when you need it.
Basic Protection
Start here
- SSH brute-force detection
- Credential stuffing detection
- Real-time monitoring
- Full audit trail (JSONL)
- Telegram notifications
- Dry-run mode by default
Automated Defense
Enable when ready
- Auto-block IPs (ufw, iptables, nftables, pf)
- AI-powered confidence scoring
- Policy-gated execution
- Cloudflare WAF integration
- AbuseIPDB reporting
- nginx rate limiting
Advanced
Optional power-ups
- SSH honeypot trap
- AI command validation (check-command API)
- Port scan and web scanner detection
- Docker container isolation
- File integrity monitoring
- Fake bot detection (rDNS)
Not sure what to enable? One command figures it out.
Scans your server, detects what's running (SSH, Docker, nginx, Suricata, osquery...), and tells you exactly what to enable.
Built for trust
What if it makes a mistake?
Dry-run by default
Nothing changes until you explicitly enable live mode.
Rust controls execution
The AI can analyze, but never execute. A compromised model cannot harm your server.
Hash-chained audit trail
SHA-256 tamper-evident chain. Every decision, every admin action. GDPR export and erase built in.
Policy-gated actions
Confidence thresholds, action whitelist, TTL cleanup. You set the rules.
Telegram approval
High-risk actions require manual approve/deny from your phone.
Compliance
Designed for regulated environments.
Inner Warden ships with the controls auditors look for. Hash-chained audit trails, admin action logging, session management, configurable retention, and GDPR data subject commands. No add-ons, no enterprise tier. Built in from day one.
SHA-256 hash-chained audit trail
Every decision and admin action recorded in tamper-evident JSONL with cryptographic integrity. Modify one entry and the chain breaks.
Admin action logging
Every capability enable, config change, IP block, and login recorded with operator identity, timestamp, and parameters.
GDPR export and erase
innerwarden gdpr export and innerwarden gdpr erase. Search by IP or username across all data files. Hash chains recomputed after erasure.
Configurable retention
Events: 7 days. Incidents: 30 days. Audit trail: 90 days. All configurable per your compliance requirements.
Session-based authentication
Dashboard auth with session tokens, configurable timeout, max concurrent sessions, and automatic expiry. Login and logout audited.
Forensic evidence capture
Automatic process state snapshots for high-severity incidents: open files, network connections, memory maps. Credentials redacted.
Designed to meet ISO 27001 Annex A controls (A.9, A.10, A.12, A.16, A.18). Read the full privacy and data protection documentation.
Safe to install
Yes, it's curl | bash. Here's what it does.
curl -fsSL https://innerwarden.com/install | sudo bash- Downloads a single SHA-256 verified binary (~7 MB) to /usr/local/bin
- Creates a dedicated service user with minimal permissions
- Starts sensor + agent via systemd (Linux) or launchd (macOS)
- Begins in dry-run mode. No firewall changes until you say so.
Want to read the script first? innerwarden.com/install
Your phone is the dashboard
Talk to your server on Telegram.
Real-time alerts, approve/deny actions, and a conversational bot that answers questions about your server's security.
Set up in one command
innerwarden notify telegramInteractive wizard. Bot token and chat ID via prompts.
Collaborative defense
Attack one node. Protect them all.
Inner Warden nodes form a mesh network. When one detects a threat, all others block the attacker automatically. Like birds that fly when one hears danger.
🔏
Ed25519 signed
Every threat signal is cryptographically signed. Tampered signals are rejected instantly.
🎯
Game-theory trust
Tit-for-tat reputation. New peers start skeptical. Trust grows with confirmed signals, drops with false ones. 3:1 asymmetry.
⏱️
Staged with TTL
No signal causes permanent action. Everything is scored and staged. Blocks auto-revert if not confirmed locally.
🛡️
Sybil resistant
New nodes start at trust 0.1. Rate-limited to 50 signals/hour. Malicious nodes get quarantined automatically.
innerwarden mesh add-peer https://peer-server:8790That's it. Identity generated automatically. Peers discovered via ping. Trust builds over time as signals get confirmed.
Ready?
Your server should defend itself. Let it.
38 kernel hooks. 48 detectors. 23 correlation rules. 8 kill chain patterns. Install in 10 seconds. Dry-run by default.