pentest-ai | Autonomous AI Pentesting CLI

the difference

Scanners flag findings. pentest-ai weaponizes them.

Burp, Nessus, and Nuclei give you a flat list of issues. We connect them into multi-step attack paths, score every chain, and validate each step with a safe proof-of-concept.

scanner output · 47 findings flat

6.1 highest single CVSSnoise · manual triage required

MED/api/proxy?url= · external fetch5.4

MEDIMDSv1 metadata reachable6.1

LOWIAM node-role over-privileged4.3

HIGHaws-auth ConfigMap: cluster-admin7.8

MEDsecrets readable in prod ns5.9

→triage queue: 47 tickets~3 days

→no chain context · no PoC · no blast radius

pentest-ai chain · CRITICAL 9.8 chained

9.8 chain severitysignal · safe PoC per step

01SSRF: /api/proxy?url=entry✓ PoC

02→ IMDSv1 · role STS credspivot✓ PoC

03→ assume node role · eks-nodelateral✓ PoC

04→ aws-auth: cluster-adminescalate✓ PoC

05→ kubectl * · prod compromiseobjective✓ PoC

→report: 1 chain, 5 validated steps~4 min

→SARIF + Sigma + MITRE ATT&CK map shipped

capabilities

Everything runs locally.
You own every byte.

MCP server exposing 32 tools to any AI client. 11 specialist agents wrap 191 security tools, share recon context, and stamp every finding with a CVSS v3.1 score and a validated PoC.

tools_per_engagement

Security tooling, wrapped

nmap · nuclei · ffuf · sqlmap · trivy · kube-hunter · BloodHound · impacket · and 183 more, all via one MCP endpoint.

web

network

binary

osint

cloud

password

11_specialists

Agents with context sharing

Each agent streams findings to the shared engagement graph. No duplicate work, no lost signal.

recon-agent scanning subdomains

web-hunter testing 47 endpoints

cloud-agent enumerating IAM

ad-attacker mapping ACLs

exploit-chainer building paths

poc-validator confirming findings

chain_engine

Templates

Six battle-tested attack paths, rotating live:

web→cloud

AD→DA

K8s→lateral

supply→prod

CI→prod

bounty→pwn

proof_of_concept

Every finding ships a safe PoC

Non-destructive reproducers, captured HAR, screenshot, request/response trace. False positives get filtered before your report.

[poc] blind SQLi confirmed · time-based · sleep(5) verified 3×

[poc] safe bucket listing · no mutation, 12 objects enumerated

[poc] XSS fires in sandbox · screenshot captured → evidence/f-014.png

ci_cd_native

Ships SARIF + JUnit + PDF

Drop pentest-ai into GitHub Actions. Breaks the build on severity gate. Posts findings as PR comments.

pentest-ai[bot] commented feat/payments-api · #482

critJWT alg=none accepted on /api/v2/admin

critIDOR on /api/v2/users/{id}/billing

poc2 findings validated, 0 false positives

✕severity gate: failed · build blocked · sarif.json uploaded

detection_output

Blue-team ready

Auto-generates Sigma, Splunk SPL, and KQL for every offensive technique used during the engagement.

Sigma

Splunk SPL

KQL

llm_red_team

OWASP LLM Top 10

Prompt injection, training-data leakage, insecure output, model DoS, covered as first-class assessment targets.

✓LLM01 · Prompt injection

✓LLM02 · Insecure output

✓LLM03 · Training data leak

✓LLM04 · Model DoS

✓LLM05 · Supply chain

✓LLM06 · Sensitive info

✓LLM07 · Insecure plugins

✓LLM08 · Excessive agency

✓LLM09 · Overreliance

✓LLM10 · Model theft

data_sovereignty

Local-first, zero telemetry

Your engagement never leaves your machine. MIT licensed. Self-hosted. Deterministic.

outbound0 B

telemetrydisabled

storagelocal only

licenseMIT

pricing

Three ways to use pentest-ai

Free OSS for individuals. Enterprise dashboard for teams. Managed Assessment delivered.

✓no per-seat ✓no per-scan ✓no hidden fees ✓cancel anytime

open_source

for builders & CI

The CLI

Free · forever

Full CLI + MCP server, no auth, no limits

191 security tools
11 specialist agents
Autonomous exploit chaining
PoC validation
CVSS v3.1 + MITRE ATT&CK mapping
SARIF + JUnit + PDF reports
CI/CD pipeline mode
Checkpoint + resume
MIT license

view on github → view on pypi↗

enterprise · most popular

for security teams

Team dashboard

$499 /mo

Hosted platform for security teams

Everything in Open Source
Shared team workspaces
SSO / OIDC / SAML
Scheduled + recurring scans
Executive PDF reports
MITRE ATT&CK coverage dashboard
Attack surface monitoring
REST API + webhooks
Jira / Slack / GitHub integrations
99.9% uptime SLA · audit log

contact sales → or see a live run first→

managed · limited

one-time engagements

Full assessment

$9,500 · one-time

typical human pentest: $15k–$30k

Full pentest engagement, delivered for you

capacity5 / quarter

by application · ~3-week turnaround

includes 3 months Enterprise ($1,497 value)

Complete autonomous pentest
Pre-engagement scoping
Exploit chain validation + PoCs
Executive + technical reports
Compliance framework mapping
Remediation priorities
30-min findings walkthrough
90-day retest window
Dedicated Slack channel

book assessment → or book a 15-min scoping call→

faq 12 questions

Common questions

productIs the CLI really free?+

Yes. All 191 tools, 11 agents, 32 MCP tools, exploit chaining, PoC validation, CVSS, SARIF + JUnit + PDF export, CI/CD mode, compliance mapping, Sigma/SPL/KQL generation, checkpoint-resume. Free under MIT. Enterprise ($499/mo) is a separate hosted dashboard for teams that want SSO, shared workspaces, and scheduling.

productWhat is autonomous exploit chaining?+

We take isolated low-severity findings and connect them into multi-step attack paths. Info disclosure + weak permission + credential reuse = Domain Admin. Each step requires your approval. Six templates cover web, AD, cloud, containers, supply chain, and API attacks.

productHow is this different from Burp, Nessus, or Nuclei?+

Scanners output a flat list of isolated findings. pentest-ai connects low-severity items into multi-step chains (SSRF → IMDS → cluster-admin), scores the chain end-to-end, and validates each step with a safe PoC. You get one real attack path instead of 47 tickets to triage.

productCan I use it for bug bounty?+

Yes, within the program's scope and rules. Use --scope to constrain targets. The bug-bounty preset auto-prunes out-of-scope hosts, disables destructive modules, and formats findings to match most program templates.

securityDoes it attack without approval?+

No. Every risky command runs in human-in-the-loop mode. You see the full command and approve or deny before execution. Set --auto at your own risk for sandboxed targets.

securityDoes my scan data leave my machine?+

The CLI runs on your machine. Scans, findings, evidence, and reports stay local unless you upload them yourself. The only exception is the LLM call: prompts go to whatever model backend you choose. Use a local model via Ollama to keep everything offline. Enterprise dashboard sync is opt-in per workspace.

securityDo I need an OpenAI or Anthropic API key?+

You need a model backend. Options: Claude, OpenAI, or a local model via Ollama (Llama 3, Mistral, Qwen). The tools themselves are free. Running with Ollama means zero cloud calls and no API key required. Set the backend with PTAI_MODEL=ollama/llama3 or similar.

securityHow do you prove findings are real?+

Every finding ships with the exact request and response, a screenshot or captured payload where applicable, CWE + CVSS scoring, and for chains, the PoC script that validated each step. Evidence is hashed and stored alongside the finding so you can replay it or attach it to a report.

pricingWhat's in the Managed Assessment?+

A full pentest engagement delivered at $9,500 one-time. Scoping, autonomous pentest, exploit-chain validation with PoCs, executive + technical reports, 30-minute findings walkthrough, 90-day retest, 3 months of Enterprise. Capacity is 5 engagements per quarter, by application.

pricingCan I try Enterprise before paying?+

The OSS CLI is the trial. Same engine, same tools, same chaining, runs locally for free. Enterprise adds shared workspaces, SSO, scheduling, API access, and the dashboard. If you want a 30-min walkthrough before committing, email sales and we'll set one up.

supportWhich OS does it run on?+

Linux and macOS are first-class. Windows works via WSL2. Install with pip install ptai, pipx install ptai, or uvx ptai. Python 3.10 or newer.

supportHow do I get help?+

GitHub Issues for bugs and feature requests. Email [email protected] for Enterprise and Managed customers. Enterprise includes a dedicated Slack channel and a 24h response SLA during business hours.

Find it. Chain it. Prove it.

Scanners flag findings. pentest-ai weaponizes them.

Every attack chain, drawn for you

Watch a real engagement in motion

Everything runs locally.
You own every byte.

Security tooling, wrapped

Agents with context sharing

Templates

Every finding ships a safe PoC

Ships SARIF + JUnit + PDF

Blue-team ready

OWASP LLM Top 10

Local-first, zero telemetry

Three ways to use pentest-ai

Common questions

Start finding what scanners miss.

Find it. Chain it. Prove it.

Scanners flag findings. pentest-ai weaponizes them.

Every attack chain, drawn for you

Watch a real engagement in motion

Everything runs locally.You own every byte.

Security tooling, wrapped

Agents with context sharing

Templates

Every finding ships a safe PoC

Ships SARIF + JUnit + PDF

Blue-team ready

OWASP LLM Top 10

Local-first, zero telemetry

Three ways to use pentest-ai

Common questions

Start finding what scanners miss.

Everything runs locally.
You own every byte.