Agents overview

A Pentestas agent runs inside your environment (a workstation, a container in your VPC, a Windows laptop) and connects outbound to the platform. Scan traffic originates from the agent, so it can reach:

Corporate intranet apps on 10.x.x.x / 192.168.x.x
On-prem GitLab / Jira / Jenkins / admin panels
Staging environments behind VPN
Windows AD / Samba / RDP services
Any target that's firewalled off from the public internet

Agents are tenant-scoped — each agent belongs to exactly one tenant, authenticates via a pa_... key, and is further locked to a pre-declared IP allowlist.

The variants

Variant	Platform	Docs
Linux agent	Debian / Ubuntu / RHEL / Fedora (also macOS via the Linux installer)	Linux agent
Windows agent	Windows 10 1809+ / Windows 11 (x64 + ARM64)	Windows agent

The Windows agent is a single self-contained .exe — native .NET 8 WPF, embeds WebView2 for scan-as-you-browse. Captures every request the user makes and streams it to the platform for active testing. No proxy, no CA cert dance.

The Linux agent is a systemd-managed service: curl … | sudo bash installs to /opt/pentestas-agent and registers a unit. It supports network scanning of internal subnets, server probes, and an optional mitmproxy + browser launcher for capture sessions on a workstation host.

Capability matrix

Capability	Linux agent	Windows agent
Server / network scan	✓	—
Web app scan (full engine in Docker)	✓	—
Browser capture	✓ (Firefox / Chrome via mitmproxy)	✓ (WebView2 + Chrome DevTools Protocol)
Runs as service	systemd	User-driven WPF app (no service)
Single binary install	No (tarball + Python runtime)	Yes (~35 MB self-contained `.exe`)
Cert trust dance	profile CA + certutil	None — uses CDP, not MITM

Install

Each agent has a distinct install flow — pick the right page above. All start with the agent key minted in Settings → Agents → New.

Security model

Tenant scope — the agent_key belongs to exactly one tenant. A stolen key gets you nothing in other tenants.
IP allowlist — the WebSocket connection is only accepted from pre-declared IPs. Admins can edit the allowlist in the UI without rotating keys.
Disable / delete — an admin can immediately disconnect a running agent. The backend drops the live socket and blacklists the key.
At-rest encryption — agent keys are stored DPAPI-encrypted (Windows) or chmod-0600 env file (Linux).
Forward-only traffic — the agent only opens outbound connections (wss://app.pentestas.com:443). No inbound ports to manage.

What the agent does not do

Store scan data locally — findings are streamed to the platform and persisted there. The agent keeps no long-term state beyond its config file.
Accept commands from anyone but the platform — the WebSocket is one-to-one with app.pentestas.com. No local socket / RPC / shell access.
Run as root unless explicitly configured — the Linux systemd unit defaults to a dedicated pentestas user with ProtectSystem=strict.