Pre-launch10 scan slots open

From Finding to Proof

For bug bounty hunters, pentesters and product security teams who have to defend every finding. Start in seconds. Investigate in minutes. Keep the source, runtime evidence and client report together.

Less time navigating tools.

More time proving vulnerabilities.

SourcePathSinkRuntimeDecisionFix
See Basilisk Run

Local-first · Authorized testing only · Every tool call audited

Evidence thread

Connected

The difference

A severity score alone does not prove a finding. Proof is a marked source-to-device chain, or a clear record that the device blocked the test.

Next-generation autonomous mobile pentesting agentBasilisk / agent runtime

01 / Basilisk

Basilisk is not another chat window bolted onto a scanner. It is the autonomous mobile pentesting agent that plans the work, delegates to specialists, calls the toolchain and keeps every decision attached to evidence. Turn a hypothesis into a structured exploit plan in seconds; runtime validation stays approval-gated and device-aware.
basiliskonline
read-only delegation · illustrative workspace

you

Prove whether this deeplink can reach the privileged screen. If it is real, show the chain.

basilisk

I will trace the entry point, verify the sink, check the device path and package the evidence. Starting with the manifest and taint flow.

static.reconentrypoint found · 3 hops
specialist.taint_flowsource + sink linked
device.validateapproval required
Approval gate

Run the device check against the connected test build?

ApproveHold
basilisk is waiting for approval…evidence preserved

The autonomous loop

1Plan
2Delegate
3Execute
4Prove

Full assessment

Static + runtime review and evidence-backed attack-chain proof.

Secret trace

Find secrets, trace their uses and confirm exposure without printing values.

Traffic + pinning

Inspect pinning, capture authorized traffic and hunt tokens/PII.

Components + PoC

Sweep exported components and deep links, then build a PoC for confirmed paths.

Specialists are read-only. Device writes and PoC execution stay approval-gated.

Without a workbench

Five tools. Five contexts.

adb, JADX, apktool, Frida and proxy. Notes and report rebuilt after the fact.

With MobLizer

One workbench. One evidence chain.

Source, runtime state, decision and report stay attached to the same finding.

02 / Early-access offer

Ten Teams. Five Free Scans.

Bring one mobile app. We will contact you directly to scope the first scan, confirm authorization and set up the evidence run.

10

early-access teams

5

free scans each

1:1

onboarding and scope

Book a scan

Leave your work email. We will reply with the artifact requirements, authorization scope and device details.

No artifact upload at signup Authorized testing only Direct email follow-up
Prefer the general waitlist? Join here

03 / The whole run

Six product surfaces, one evidence thread. Follow any tile into the deeper workspace without losing the finding.

Introducing

Basilisk

Autonomous operator

Plans the work, delegates to specialists, asks before touching a device and keeps every decision attached to evidence.

static.reconentrypoint found
specialist.taint_flowsource + sink linked
device.validateapproval required
evidence.packagereport ready
Open the autonomous run

Evidence chain

Source, path and sink in one view.

EntryActivity.java

42Uri route = intent.getData();

47startActivity(launch);

3 hops, source-linked, no sanitizer on the path.

Full Audit

20

roles in one panel

recon-triagedone
manifestdone
storage-cryptodone
webview-securitydone
taint-flowdone

Confirmed means the panel could not refute the evidence.

Runtime evidence

See what the app does.

DynamicNetworkFrida

Static hypothesis, runtime proof or a visible blocked state.

Investigation

One finding, one workspace.

High

Deeplink reaches a privileged intent

path only · 55% confidence

Evidence, analysis, validation and history stay replayable.

Client report

A report the client can audit.

target.app · Penetration Test Report

source · runtime · decision · remediation

HTML, Markdown, PDF, DOCX, XLSX, SARIF, JSON and CycloneDX exports carry the same evidence.

Swipe to explore the run.

Illustrative workspace. The outputs reflect the product model, not a live customer run.

04 / Full Audit

20 roles examine the same artifact, challenge each other's claims, and keep the coverage gaps visible. Confirmed means the evidence survived the panel, not that a model sounded confident.

The evidence bench

A candidate only becomes confirmed when the runtime path is supported and the adversarial panel fails to refute it. A blocked device is reported as blocked.

  1. 1

    Recon

    Attack-surface map and target context.

  2. 2

    Specialists

    Manifest, storage, WebView, network, native and more.

  3. 3

    Runtime

    Device-backed verification where the target allows it.

  4. 4

    Panel

    Prosecutor plus two independent skeptics.

  5. 5

    Report

    Evidence bundle and client-ready HTML.

Panel standard

Prosecutor + two skeptics. A confident paragraph never counts as a verdict.

target.app

full audit · same artifact · same evidence

panel complete

20

agent roles

28

candidates

9

confirmed

Agent roster

18 complete · 2 skipped

recon-triage

attack surface

manifest

components · deep links

storage-crypto

keystore · files

webview-security

bridges · CSP

network

pinning · traffic

taint-flow

source → sink

native

ELF · JNI

resilience

root · debug

Pipeline

Recondone
Specialistsdone
Runtimedone
Panelnow
Reportdone

Panel verdict

Confirmed

Source + sink verified. Runtime evidence attached. No refutation found.

Known evidence stays visible

Marked code, runtime artifacts and the report are linked.

Gaps stay honest

Missing devices, source and validation are explicit.

05 / One finding, end to end

The same finding moves through code, path, runtime, decision and report. Switch the lens without losing the evidence.

One finding · five views

High severity

Deeplink reaches a privileged intent

confirmed

A crafted deeplink crosses three hops and reaches a privileged intent. The evidence stays attached to the source, the path and the device prerequisite.

Target
target.app
Surface
Exported activity
Evidence
Source + sink
Runtime
Approval required

Confirmed means the evidence survived the panel. A blocked device remains a visible gap.

41Uri route = intent == null ? null : intent.getData();
42if (route == null) return;
43String next = route.getQueryParameter("next");source
44if (next != null) {
45 Intent launch = new Intent(this, TargetActivity.class);
46 launch.putExtra("url", next);
47 startActivity(launch);
48}
01The deeplink parameter enters through the exported activity. evidence preserved

06 / Runtime is evidence

Static analysis creates the hypothesis. Choose dynamic analysis, network, Frida or the investigation workspace. Each view keeps the same finding in context.
target.app · live connected
SELinuxenforcing
package installedyes
start activityforeground

Captured signal

Deeplink accepted by the exported entry point.

runtime evidence attached to finding

runtime.trace

14:22:09 activity.start target.app

14:22:10 intent.received EntryActivity

14:22:10 uri.parameter next=...target

14:22:11 boundary.privileged true

14:22:11 evidence.capture runtime_01.json

evidence.summary

finding deeplink → privileged intent

status observed

artifact runtime_01.json

Dynamic analysis

Target-scoped runtime checks with visible preconditions and evidence capture.

07 / The speed of proof

Start in seconds. Let the audit run in minutes. Days of tool-hopping become one evidence chain, with the source, runtime status and client report attached.

Seconds

Start an investigation

No upload queue or cloud round trip before the run begins.

33 min

Fastest stored audit

18-role audit; the second stored run completed in 41 minutes.

139

Detection rules

82 behaviour rules and 51 deterministic attack chains.

57

Agent tools

14 Android and 21 iOS Frida scripts in the registry.

Built for people who have to prove it

Bug BountyIndependent PentesterSecurity ConsultancyProduct SecurityEnterprise AppSec

The audit roster is 20 roles; 18 roles run on a given target. The 33 and 41 minute runs are validation-corpus results with no device attached, not universal customer benchmarks.

08 / The business case

Model your own team and assumptions. The result is only as honest as the inputs. These defaults are illustrative, not a measured MobLizer claim.

Your assumptions

Formula: (manual analysis + manual reporting − assisted analysis − assisted reporting) × assessments × 12 × hourly cost.

Illustrative annual result

1008 hours

saved across the modeled year

Annual labor value

$120,960

Per assessment

21 hours

Net annual value

$120,960

ROI

Add a platform cost

This calculator does not claim measured savings. Validate the assumptions with your own benchmark before using the result in a business case.

09 / Local-first, by architecture

The backend binds 127.0.0.1. Your device, ADB and Frida server stay on your desk and LAN. The only bytes that ever leave are opt-in calls you trigger.

Settings · Privacy

Local-only mode

Flip it and MobLizer refuses cloud AI entirely. A persistent LOCAL / CLOUD indicator always shows the current mode.

Bring your own model: 7 providers

  • Ollamalocal
  • OpenAI
  • Anthropic
  • OpenRouter
  • DeepSeek
  • ScaleMax
  • Kimi

What stays on this machine

  • Device traffic

    ADB and Frida sessions stay on your desk and LAN.

  • Artifacts

    APKs, manifests and decompiled files remain local.

  • Evidence

    Findings, traces and client reports stay in your workspace.

Your workstationADB · Frida · JADX
blocked
Cloud AIrefused

Local-only mode refuses every cloud provider. Ollama runs entirely on your machine.

The runner allowlist: 9 programs, nothing else

  • adbLOW
  • fridaMEDIUM
  • jadxLOW
  • apktoolLOW
  • apksignerLOW
  • keytoolLOW
  • javaLOW
  • mitmdumpMEDIUM
  • gitMEDIUM

Never a raw shell. Args are passed as a list, every call is timeout-bounded and audited, and HIGH-risk actions (like installing a system CA) need explicit acknowledgement.

10 / Client-ready, and iOS without a Mac

Findings carry business impact, remediation, PoC and evidence where they apply, mapped to CWE, OWASP Mobile Top 10, MASVS, with a reproducibility appendix. Export MD, HTML, PDF, DOCX, XLSX, SARIF, JSON, CycloneDX.

Business impact

Every finding leads with plain-language impact, what an attacker gains and why it matters to the business, so a report reads without a translator.

MDHTMLPDFDOCXXLSXSARIFJSONCycloneDX
iOSDamn_VulnBank · app-4.ipa

IPAs analyse statically on any OS: Mach-O protections, entitlements, ATS, URL schemes and built-in Objective-C class metadata, static analysis only, no runtime or keychain dumping. FairPlay-encrypted binaries are flagged; plist, entitlement and framework checks still run.

7
URL schemes
70
Frameworks
12
Usage strings
2
Secrets
12
Endpoints
0
Cleartext

No macOS required. Works on any IPA.

Before you ask

The fastest way to lose technical trust is to imply more automation than the product actually provides.
01Does MobLizer upload my APK or IPA?

The workbench is local-first. Artifacts and tool execution stay on your machine; cloud AI is opt-in and can be replaced with local Ollama.

02Does Basilisk execute every action automatically?

No. Interactive workflows pause for medium and high-risk tools, runtime validation requires a device, and destructive commands are blocked. A full audit still requires explicit authorization before it starts.

03Does iOS have the same runtime support as Android?

No. iOS static analysis covers Info.plist, entitlements, Mach-O hardening, frameworks, secrets and attack surface. Runtime Frida and device-backed validation require a jailbroken device and SSH. Android has the deeper dynamic and PoC path today.

04What does a full audit include?

A 20-role roster with 18 active roles on a given target, specialist review, an adversarial panel, source-linked findings and an explicit evidence-gap section. Confirmation requires runtime evidence; a missing device is reported, not guessed.

05Can I use it offline?

Local analysis and tool execution are designed for the workstation. Local AI can run through Ollama. Cloud model use is a separate, visible choice.

06What reports can I export?

Audit reports ship as HTML and Markdown. App reports support HTML, Markdown, PDF, DOCX, XLSX, SARIF, JSON and CycloneDX.

07What happens after I book a scan?

We reply directly with scope, authorization requirements, artifact details and device prerequisites. The first 10 accepted teams receive five free scans each.

08Is MobLizer available today?

MobLizer is pre-launch. Join the early-access list for launch updates and benchmark results.

The mobile attack surface

MobLizer is pre-launch. Join the waitlist for early access, one launch email, no spam.

Local-first · CWE / OWASP Mobile Top 10 / MASVS · pre-launch