Web App · ATS — SENTRA AI Threat Shield
Datasheet for the AI web application firewall that protects websites and APIs (version 2.26).
Version: 2.26 · Layer: Enforce · Type: AI web application firewall (inspecting reverse proxy)
An AI web application firewall that sits in front of a website and inspects every request before it reaches the server. Injection, credential stuffing, scrapers, AI agent swarms and floods are stopped at the edge, with no change to the application.
Key capabilities#
Inspection#
- Five checks on every request — network, fingerprint, behaviour, AI intent and policy — scored together into one verdict. Clear attacks are blocked, likely ones challenged (browser check), everyone else passes straight through. The challenge and block lines are adjustable, with Balanced, Strict and Paranoid strategies.
- Monitor and Enforce modes — Monitor records what would be stopped; Enforce blocks, challenges and bans.
- AI agents and swarms — the good-agent passport verifies genuine AI agents and crawlers by signature and network and catches impostors; swarms that spread one job across many addresses are grouped into one campaign and contained.
- Agent maze — bots that keep coming back after being blocked get believable decoy pages instead of an error.
- Virtual patches for specific, actively exploited CVEs, with coverage reporting against the CISA Known Exploited Vulnerabilities list.
- Breached-password check at sign-in (k-anonymity; the password never leaves the box).
- Rate limits, tarpit and automatic bans; allow and block lists by address, network, country and ASN.
Site protection#
- Page Shield — watches the scripts your pages load and alerts on new or injected ones.
- API Shield — validates content types, sizes, tokens and JSON fields per API route.
- Response hardening — adds missing security headers, cookie flags and HSTS; strips version banners.
- Websites and certificates (edge) — add websites, automatic Let's Encrypt certificates, per-site inspect or pass-through, emergency bypass.
- Fail-open guard (edge) — if the shield stops answering, traffic goes straight to the web server within seconds and an alert is raised.
Intelligence (module sentra-ids)#
- IDS hub — Suricata and Zeek network sensors in one real-time view.
- Threat feeds — open IDS signature sets and IP reputation lists, updated automatically.
AI analyst & Copilot (module sentra-ai)#
- AI second opinions on traffic, AI assessments and the Copilot assistant.
- Works with OpenAI-compatible providers, including private on-premises models (Ollama, vLLM, LM Studio, NIM).
- Cross-model disagreement telemetry flags adversarial probing.
Operations#
- Overview dashboard, customisable per user (widgets, order, size) with a default layout for everyone; Live view; wall mode for NOC screens.
- Incidents and campaigns, events, who (visitor investigation), bans.
- Policy with staged changes, deploy and roll back to earlier revisions (deterministic policy compiler).
- Reports — report builder with sections, periods and filters; schedules by e-mail; archive; print/PDF and CSV. Value report with a clearly labelled estimate of what was saved.
- Tools — passive security Scanner, Bot Scorecard, and Active test (optional assessment tools: nmap, nikto, gobuster, nuclei, whatweb) for sites you own.
- Security drills to rehearse attacks safely.
- LAN storm guard — detects broadcast, multicast and ARP storms and switching loops.
- Alerts in plain language by Telegram and WhatsApp (English, Bahasa Melayu or both), daily and monthly summaries; e-mail (SMTP) for scheduled reports.
Deployment#
| Option | Description |
|---|---|
| Same server | Beside the website, forwarding over loopback. Isolated service user, CPU and memory caps. For a single site and trials. |
Dedicated edge (install --edge) |
Its own server holds ports 80/443 and the certificates; web servers sit behind it on private addresses. Recommended for production. |
| GUARD node | Provisioned and managed from SENTRA GUARD through the node agent; each customer gets their own firewall. |
Requirements#
| Item | Requirement |
|---|---|
| Operating system | Linux with systemd; Debian/Ubuntu (apt) for the edge install (certbot, nftables) |
| Python | 3.8 or newer |
| CPU / RAM / disk | Sized per deployment — contact SENTRA. The service runs within a memory cap of 768 MB (same server, 60% CPU cap) or 1.5 GB (edge). |
| Network | Edge: public ports 80/443. Outbound HTTPS for licence check-in, release checks and any feeds or AI providers you enable. |
Ports and protocols#
| Port | Use |
|---|---|
| 80, 443 (TCP) | Public websites (edge install) — HTTP redirect and HTTPS |
| 9280 (TCP) | Dashboard — bound to 127.0.0.1 by default; publish opens it on the private IP; optional HTTPS (dashtls on) |
| 9281, 9283 (TCP) | Demo site and SENTRA website lanes on a non-edge install (local by default) |
| 443 (outbound) | Licensing check-in, release server, feeds, AI providers, alert channels — only those you enable |
Integrations#
Telegram, WhatsApp (Meta templates), SMTP e-mail; Suricata, Zeek; OpenAI-compatible AI providers and private models; S3-compatible storage, SFTP or a folder for off-box backups; Let's Encrypt; SENTRA GUARD (node agent and central identity).
Licensing#
| Module | Covers |
|---|---|
sentra |
The shield: inspection, Policy Fabric, bans, incidents, reports, dashboard |
sentra-ids |
Network sensors & threat feeds: Suricata, Zeek, threat feeds and signature updates |
sentra-ai |
AI analyst & Copilot |
No user or site count. 30 days of grace for an unlicensed module, then restricted: protection continues, changes for that module pause.
Management and security#
- Users with fully customisable roles (15 permissions in four groups; built-in Master Admin), two-factor sign-in, sign-in rules, and a glass-break account for when the managing GUARD is unreachable.
- Central identity from SENTRA GUARD when managed: GUARD users sign in with their GUARD accounts, mapped to local roles.
- Tamper-evident audit ledger of security decisions and admin actions.
- Backups nightly, with optional encrypted off-box destinations; restore from the console or command line.
- Signed updates from the release server, installed only when an admin chooses.