PentAGI
pentagi.com- Category
- Cybersecurity
- Rank
- No. 2177Tools index
Previous survey · No. 3165 ·
- Listed in
- #3 Review code for security
- Pricing
- Paid
- Type
- TOOL
- Use case
- Security & Identity
- Interfaces
- Web
- GitHub
- 33.7k stars
- Latest release
- v2.2.0
- Date
About
An MIT-licensed, self-hosted platform for authorised penetration-testing automation. Model-driven agents plan tasks, run tools in Docker, retain results and prepare reports. It requires deployment and network controls; no finding or containment guarantee follows from autonomy.
At a glance
PentAGI is a self-hosted platform that lets model-driven agents plan and run penetration-testing tools, retain results and produce a report. It is worth evaluating for security teams that want an inspectable automation platform and can operate its infrastructure. It is a substantial deployment with powerful execution privileges, not a low-maintenance website checker or proof that a system is secure.
Good fit
Security engineers experimenting on systems they own or are explicitly authorised to test. A flow connects a goal to planning, command execution, memory and reporting, so you can examine how a finding was reached instead of receiving only a generated answer. The platform supports cloud or local models; model choice, test scope and the operator’s review still determine whether the output is useful.
Consider something else when
You cannot maintain a Docker service, control its network access or review what an autonomous testing agent does. The backend manages Docker and runs as root according to the security documentation. Sandboxed tools are a design boundary to configure, not complete isolation from every consequence: callback ports and outbound access need attention, and a dedicated worker is recommended for production.
Start here
Plan an isolated lab and a deliberately vulnerable target under your control before deploying. The documented console is https://localhost:8443; the first useful check is inspecting the generated plan, command log and final report for one narrow lab task. Change the initial administrator password and configure random secrets before exposing the service. Do not begin with an unrestricted instruction to test a real organisation. No service, scan or exploit was run for this guide.
Cost and requirements
The open-source edition is MIT-licensed. The installation site lists Docker Compose, 2 vCPU, 4 GB RAM, 25 GB disk and access to a model provider; local-model hardware, hosting and paid model/search usage are additional costs. Optional observability and graph services add resources. Enterprise adds delivery and management features under separate commercial terms. Flow records stay in the deployment, while selected cloud models and search engines receive relevant inputs.
Limits to know
Self-hosting alone does not keep all traffic local. Documentation describes default DuckDuckGo search, a server update/usage-summary request disabled by UPDATE_CHECK_INTERVAL=0, and separate installer reporting unaffected by that switch. The proxy setting does not cover sandbox or browser traffic. Deleting a flow leaves its data directory on disk, according to the README; operators must manage that cleanup. Source review covered selected Docker/terminal code and deployment configuration, not containment or vulnerability-detection accuracy. No installer, image, model, target interaction or benchmark was executed.
Sources checked 2026-11-05. Source-based assessment.
- pentagi.com
- pentagi.com/docs/configuration/security/
- pentagi.com/docs/deployment/worker-node/
- pentagi.com/get-started/
- github.com/vxcontrol/pentagi/blob/main/README.md
- github.com/vxcontrol/pentagi/blob/main/LICENSE
- github.com/vxcontrol/pentagi/blob/main/backend/go.mod
- github.com/vxcontrol/pentagi/blob/main/docker-compose.yml
- github.com/vxcontrol/pentagi/blob/main/backend/pkg/tools/terminal.go
- github.com/vxcontrol/pentagi/blob/main/backend/pkg/docker/client.go
What it can do
Scan network for open ports and services
Target IP address or network range → List of open ports and running services
Identify system vulnerabilities
Target system or application → Vulnerability assessment report with CVE details
Perform automated exploit attempts
Identified vulnerabilities and target system → Exploitation results and proof of concept
Analyze web application security
Web application URL → Web security assessment with identified weaknesses
Generate penetration testing reports
Security testing results and findings → Formatted penetration testing report with recommendations
Conduct SQL injection testing
Web forms and database endpoints → SQL injection vulnerability findings
Test for privilege escalation paths
Compromised system access → Available privilege escalation methods and paths
Why it made the leaderboard
Connects testing plans, command records and reports in an inspectable workflow for security teams. Hosting, model costs, Docker authority and review of agent actions remain the operator’s responsibility.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.