The platform

One chat window.
A full red team.

Recon, exploitation, validation, and reporting. Every layer of an enterprise pentest, delivered by a coordinated swarm of 13 AI agents and 41+ scanner modules. You drive it from chat.

Chat-driven workflow

The whole product
is a conversation.

No CLI. No Burp config. No "scope" form to fill. You sign in, type a URL into the chat, and watch Zeroday IQ run a real pentest in front of you.

An LLM intent classifier (with a regex fallback) decides whether your message is a scan request, a follow‑up question on a previous scan, or a general security question. SCAN intents trigger a real background scan and stream progress live; FOLLOWUP intents are answered by an LLM bound exclusively to your stored scan data. It physically cannot invent findings the scan did not produce.

  • Trigger a scan in plain English: "scan example.com" or "test our staging API"
  • Ask follow‑ups: "show only the criticals," "explain the SSRF," "give me a curl PoC"
  • Pick your model per request: OpenAI GPT‑4o or Claude Sonnet 4.6
  • Every scan creates a conversation thread you can return to later
chat / intent.py · classify_intent() gpt-4o-mini · temp=0
user scan staging.acme.com for OWASP Top 10
SCAN
user explain the SSRF, give me a curl PoC
FOLLOWUP
user what's the difference between SSRF and CSRF?
GENERAL
SCAN → background thread streams progress live FOLLOWUP → grounded on stored scan only
Scan pipeline

Recon to remediation, fully automated.

Point Zeroday IQ at your stack. About ten minutes later (twenty five for a deep scan), the report is on your dashboard.

STEP 01

Recon

DNS, CT logs, subdomain enumeration, OSINT, GitHub leaks, dark‑web breach data. Synthesised into an attack‑surface map.

STEP 02

Parallel wave

41+ modules dispatched in one thread‑pool wave with per‑module timeouts (Nuclei 200s, vuln‑scanner 180s, WAF‑bypass 90s; others tighter).

STEP 03

Risky-subdomain deep-scan

Up to three high‑risk subdomains (admin, staging, dev, vpn, api) auto‑deep‑scan with their own sub‑module batch.

STEP 04

Validate & report

The Validator agent drops anything without raw evidence. The Report agent ships a compliance‑mapped PDF the moment the scan completes.

Multi-agent orchestrator

A coordinated AI red team.

One set of agents runs on every external scan; another set activates when you deploy the internal network agent. Each agent owns one phase of the kill chain, with its own toolset, memory, and reasoning trace.

External agents

Recon

DNS, OSINT, CT‑logs, subdomains, GitHub leaks, dark‑web breach data. Synthesised into a target attack‑surface map.

PHASE · DISCOVERY

Enumeration

Walks endpoints, params, GraphQL introspection, JS bundles, and config exposures to build the live application surface.

PHASE · ENUMERATION

WAF detection

Fingerprints the WAF in front of the target, then runs 70+ bypass payloads to score real evasion risk before exploits run.

PHASE · WAF

Vulnerability

Drives the 41+ scanner modules: Nuclei (9000+ templates), CVE matching, custom active checks. Aggregates raw findings.

PHASE · VULN

Exploit

Confirms each candidate with a real PoC and writes the attack narrative: SSRF chains, SSTI, deserialization, BOLA, JWT misuse.

PHASE · EXPLOITATION

Validator

The zero‑FP gate. Re‑checks every finding against raw scanner evidence; anything without proof is dropped before the report.

PHASE · ZERO-FP

Report

Builds the executive summary, severity rollup, immediate‑actions list, and long‑term remediation plan. Hands off to the PDF generator and the chat formatter.

PHASE · REPORTING

Why agents, not a single LLM?

One context window cannot hold a real pentest. Each agent has its own toolset, its own memory, and its own reasoning trace. So when you click a finding, you see exactly which agent ran what command and what came back. Auditable, not a black box. The Validator is the last gate, and it has no power to add a finding. Only to drop ones that fail proof check.

REASONING TRACE · PER-AGENT EVIDENCE

Internal scan agents

Network discovery

From a host inside the perimeter, enumerates internal CIDRs, live hosts, and exposed services that never touch the internet.

INTERNAL ONLY

Active Directory

Probes for weak Kerberos config, AS‑REP‑roastable accounts, kerberoastable SPNs, and ACL paths to Domain Admin.

INTERNAL ONLY

Credential discovery

Looks for plaintext credentials in shares, configs, and process memory; cracks weak hashes; correlates leaked‑cred reuse.

INTERNAL ONLY

Lateral movement

Maps reachable services per credential (SMB, WinRM, SSH, RDP) and validates pivot paths between hosts.

INTERNAL ONLY

Privilege escalation

Per host, checks for unquoted services, SUID gaps, sudo misconfig, kernel CVEs, GPO abuse, token‑impersonation paths.

INTERNAL ONLY

Attack-path · blast-radius · MITRE

Composes findings into multi‑step kill chains, scores blast radius (which data classes are reachable), and maps every step to MITRE ATT&CK tactics & techniques.

INTERNAL ONLY
Internal & network scanning

Test what's behind the firewall.

External pentests miss internal RCE, weak shares, exposed databases, and misconfigured services that never touch the public internet. Zeroday IQ ships a lightweight Docker agent installed on a host inside your network. It runs the same 41+‑module pipeline from inside the perimeter and streams findings back to the dashboard.

  • Super‑admin gated. Every internal scan is a reviewed engagement, not a free‑for‑all.
  • Token‑scoped agent registration, signed scan results.
  • Internal findings appear in the same report alongside the external surface.
  • Per‑user usage limits and full activity log for compliance.
Internal scan · zid-internal-7f3a 10.0.0.0/16 · 7 findings
Crit
Jenkins script-console reachable without auth on 10.0.6.8
Runs as root · pivot path leads to production DB backup share · CVE-2024-23897 confirmed
High
MongoDB 4.2 exposed, auth disabled on 10.0.4.12:27017
3 databases readable · 1 contains users with password hashes · BCrypt cost-10
High
AS-REP roastable accounts found in DC 10.0.0.5
2 service accounts with DONT_REQ_PREAUTH · hashes captured · cracked in 4m offline
Med
SMB share \\10.0.7.21\backups readable anonymously
4.2 GB · contains nightly DB dumps · last modified 6 hours ago
Low
Outdated OpenSSH on 10.0.6.12 (build host)
7.4p1 · CVE-2018-15473 username enumeration · informational only
ai/analyzer.py · validator_system_prompt locked
# Immutable. Cached server-side for prompt-cache hits. SYSTEM_PROMPT = """ You are the Validator agent. You receive raw scanner findings as input. Apply these rules without exception: 1. Report ONLY findings backed by raw scanner evidence. 2. NEVER invent, infer, extrapolate, or speculate. 3. Every issue MUST include a proof field with the literal scanner output that backs it. 4. Severity follows CVSS 3.1, not narrative tone. 5. If proof is missing or ambiguous, DROP the finding. """ # Last validator pass on scan 4f2a8c… retained: 17 · hallucinated: 0 · dropped: 5
AI layer

Grounded by design.

Every model call (analysis, chat, classification) runs against a static system prompt that explicitly forbids invented findings. The variable scan data goes in as the user message; the immutable rules are cached server‑side for prompt‑cache hits and audit consistency.

  • OpenAI GPT‑4o for analysis, Claude Sonnet 4.6 fallback
  • Structured JSON output, validated before storage
  • Chat answers grounded exclusively in your stored scan
  • Prompt‑cache aware: 50% discount on the static prefix
Compliance mapping

Every finding maps to your auditor's checklist.

Hand the report to your auditor as is. Each finding is tagged with control IDs across seven frameworks plus CWE, CVSS 3.1, and MITRE ATT&CK references.

SOC 2 Type II

CC6.1, CC6.6, CC7.1 trust‑services criteria mapped per finding.

ISO 27001:2022

Annex A.8 / A.12 / A.14 / A.18 controls.

HIPAA

Security Rule §164.308 / §164.312 safeguards.

GDPR

Art. 32 technical & organisational measures.

PCI DSS 4.0

Requirements 6 (secure dev) & 11 (testing).

NIST CSF 2.0

Identify · Protect · Detect functions tagged.

OWASP Top 10 (2021)

A01 to A10 categories auto‑mapped.

CWE · CVSS · MITRE

Per‑finding CWE ID, CVSS 3.1 vector, MITRE ATT&CK technique.

Beyond unauthenticated

Past the login wall. Into the kill chain.

Authenticated surface, cross‑user access control, and ongoing change‑detection. The three layers commodity scanners either skip or charge enterprise prices to unlock.

Authenticated scans

Hand over a login URL, username, and password. Zeroday IQ runs the form login in a real browser, captures the session, and exercises the post‑auth surface. The place 80% of real bugs live. JWT and OAuth‑based auth work the same way.

FORM LOGIN · JWT · OAUTH

Cross-user BOLA detector

Provide a second account and Zeroday IQ tests whether user B can access user A's resources. IDOR / BOLA / horizontal‑privilege bugs that pure unauthenticated scanners simply cannot detect. The exact class of bug that keeps causing breaches.

IDOR · BOLA · HORIZONTAL PRIVESC

Continuous monitoring

Schedule re‑scans every 1, 6, 12, 24, 48 hours or weekly. Each run is diffed against the previous; only new findings trigger alerts. Notifications via Slack, webhook, or email. Alert‑fatigue prevention is the point.

DIFF-BASED · SLACK / WEBHOOK / EMAIL

See it run on your stack.

Sign in and chat your first scan in the next ten minutes. Your email‑domain is automatically in scope, no setup required.