Prove what's exploitable. Not what a scanner guessed.
Aegis is a verified-exposure and remediation platform. Every finding is either Verified by Aegis or Attested by a licensed human. Never accepted on an AI's say-so.
Coming Soon - Reach Out for Early AccessClick a verdict. See what it actually takes to earn it.
Every finding in Aegis carries exactly one of these three labels - never a fuzzy confidence score, never "AI thinks this is probably real."
The default verdict, and the one nearly every finding carries. A real request, a real response, a real comparison - authz_test uses two genuine accounts to prove one can reach what the other shouldn't. No LLM sits anywhere in this path. If it says Verified, a machine reproduced the exploit and captured the evidence.
evaluate_request() gates every check - scope, method, rate limit, kill switch - before a single byte goes out.
You're already buying this. Badly, and in pieces.
An "AI pentest" subscription that narrates findings instead of proving them. An open-source SAST tool with zero support. A compliance consultant just to fill out a spreadsheet. And when you need a signature, a pentest firm that takes weeks to book.
Aegis: SAST + Deterministic Pentesting + Attack Graph + Compliance, one evidence model - optional AI deep-dive, optional human attestation, cloud or on-prem or fully self-hosted.
Seven steps. Every one of them provable.
SAST, web/API, network, and cloud-posture engines run on schedule or on demand - cloud-side or through the on-prem agent.
Deterministic checks produce real proof. authz_test compares real accounts. Nothing is marked verified without evidence attached.
Findings connect automatically into attack paths - including cross-project links from a SAST finding to a live discovered endpoint.
Severity, exposure, and business criticality get multiplied by real-world signal: CISA KEV and FIRST.org EPSS, synced hourly.
Stack-aware fix templates branch by the finding's real language and context - not a generic "sanitize your input" boilerplate.
The exact original check runs again through the same safety engine. Fixed means fixed. A regression is flagged regressed, never silently reopened.
A licensed human can sign off on demand - append-only, with reviewer, scope, and evidence preserved, and a report anyone can verify is genuine.
Four exhibits. One evidence model.
Drag or scroll sideways - each exhibit is a real engine, not a mockup.
Reads your source before it ships, not just your deployed app from the outside.
Real SAST - the layer almost no "AI pentest" tool actually hasWraps the Semgrep engine against Aegis's own ruleset, not Semgrep's registry config, whose license terms restrict commercial SaaS use.
Real zip-slip, zip-bomb, and symlink defenses on the upload path, not a naive extractor.
Comments only on findings that are provably new or regressed against a baseline - never blanket noise.
Real probes with real proof - no LLM in the loop, nothing accepted on its say-so.
Replaces the guesswork in most automated pentest toolingTwo distinct real accounts, compared against each other to see what each can actually reach.
http_discovery, header_audit, and tls_audit - all deterministic, all evidence-producing.
Authorization and Cookie headers are stripped before anything is ever persisted.
Not "run the same Docker containers on your own servers." An actual outbound-only agent.
The on-prem gap almost every competitor in this category hasCGO_ENABLED=0, no runtime to patch, small enough for a security team to actually audit.
The agent always polls the control plane. The control plane never dials the agent. Full stop.
The agent genuinely verifies the control plane's identity too - not just the reverse.
Findings that talk to each other, ranked by what's actually being exploited right now.
Replaces manually stitching findings into a story yourselfIncluding cross-project links from a SAST finding to a live discovered endpoint, tenant isolation intact.
Real-world exploitation signal multiplies the base priority score, synced hourly.
A finding that regresses after being marked fixed is flagged regressed, never silently reopened.
Run it where your data has to live.
Pick a mode per project, not once for the whole company. Every mode routes through the same safety policy engine.
Point Aegis at a target and start scanning. No infrastructure of your own to run or patch - the control plane reaches your target directly over the public internet.
- Zero infrastructure to run
- Every engine available, including SAST
- Fastest path from signup to first finding
A framework mapping tied to live evidence, not a checklist.
A control with no automated evidence stays not_tested forever. Absence of evidence is never treated as evidence of absence.
Mapped per control to exactly which scanner would close the gap, with a computed verification_source: scanner-verified, human-attested, or not verified. A control with no automated evidence stays not_tested forever - absence of evidence is never treated as evidence of absence.
A paid, human-signed attestation layer on top of your report. An Aegis-staff reviewer inspects the underlying findings and signs a real certification statement with its own dedicated Ed25519 keypair - separate from every other signing key in the system.
Anyone - an auditor, a regulator, a procurement contact - can verify a certificate is genuine through a public, unauthenticated endpoint. No login, no trust-me.
Authorization and Cookie headers are stripped before evidence is ever persisted to disk - proof of exploitability without a second, self-inflicted leak.
The exact original check runs again through the same safety engine. A regression is flagged regressed, never silently reopened as something new.
PDF, Markdown, and DOCX export, with an AI-polish pass that can tighten the prose but can never invent a finding that wasn't there.
Security isn't a bullet point we added later.
evaluate_request() and its siblings for DNS, TCP, and cloud-API calls are the single place scope authorization, rate limits, testing windows, and the kill switch get enforced - cloud or on-prem, zero exceptions.
Every query is filtered by the authenticated user's organization at the router layer - deliberately explicit and testable, not a row-security policy you have to trust blindly.
Release manifests, on-prem license files, and Certified Audit Report certifications each use their own dedicated Ed25519 keypair. Compromising one can never compromise the others.
Passwords and API keys are argon2-hashed. Stored credentials use Fernet symmetric encryption with MultiFernet key-rotation support.
Org-wide and per-scan kill switches are checked at every loop iteration - verified live to reach a cancelled state within about two seconds, even mid-AI-pentest-run.
What you get with one platform instead of five.
| Aegis | AI-pentest wrappers | Open-source SAST tools | Cloud posture scanners | Traditional pentest firms | |
|---|---|---|---|---|---|
| Deterministic, evidence-backed findings - not AI narration | |||||
| Real static code analysis (SAST) of your source | |||||
| Genuine on-prem deployment - static binary, 0 inbound ports | |||||
| Automated retest that confirms a fix actually worked | |||||
| CISA KEV / FIRST EPSS real-world threat-intel priority | |||||
| Automatic attack-path correlation across findings | |||||
| Licensed human attestation available on demand | |||||
| Compliance mapping tied to live evidence, not a checklist | |||||
| Continuous / scheduled - not a one-time engagement | |||||
| Cryptographically signed, publicly verifiable audit reports |
partial support in some tools in the category - varies by vendor.
Priced by coverage, not by seat.
Every SaaS tier includes the full deterministic pipeline. Higher tiers add compliance frameworks, scan concurrency, on-prem agents, and the AI Pentester Agent. Running the whole platform on your own infrastructure is licensed and sold separately.
Starter
For a single team that needs real coverage on real targets.
- 5 concurrent scans
- 3 backup slots
- 1 on-prem agent
Team
For teams that need the AI Pentester Agent and monthly certified proof.
- 10 concurrent scans
- 10 backup slots
- 5 on-prem agents
Enterprise
For portfolios, regulated environments, and dedicated SLAs.
- 20 concurrent scans
- 50 backup slots
- 25 on-prem agents
Self-Hosted
The entire platform on your own infrastructure - licensed and sold separately from the SaaS plans.
- Signed releases, offline update verification
- Dedicated support channel
- Custom SLA
VERIFIED
Stop trusting a scanner's word for it.
Deterministic proof by default, licensed human sign-off on demand, and AI kept exactly where it belongs - bounded, optional, and never the final word.
Coming Soon - Reach Out for Early Access