Run a scan, read the verdict, hand off the report.
Xseth is a teammate, and it pays to work it like one: scope it well, read the evidence, and let it do the triage so you spend your time on judgment. Here is the end-to-end loop, plus the screens that show the agent's thinking.
Open the consoleStep by step
Confirm your authorization first
Only submit targets you own or have written authorization to test. This is non-negotiable and it is enforced — you can only scan domains on your authorized scope, the intake gate refuses known-sensitive categories outright, and every target the agent derives afterwards is re-checked against your scope before anything runs. If a target is out of scope, it is refused, with a reason.
Submit a target
Open the console, start a new scan, and enter a domain or IP. The intake gate evaluates it in real time and tells you whether it was approved or declined — and why. Once approved, the scan is queued and you can watch the recon stages progress live.
Read the recon results
When recon lands you get the raw map of your surface: the subdomains discovered, the open ports, the live web services with their detected technology, and the per-asset service versions. This is the evidence the assessment is built on — you can always trace a finding back to what was actually observed.
Work the threat assessment
The assessment is where you spend your time. Findings are ranked by severity; open one to see its evidence line, its impact, and how to remediate it. A CONFIRMED badge means a probe verified the finding is real and reachable; an unproven badge means the model flagged it but it has not been validated yet. The prioritized next steps tell you what to look at first.
Ask Thoth
Every assessment ships with Thoth — Xseth’s grounded chatbot, named for the keeper of knowledge. Ask it things like "which findings are actually confirmed?", "walk me through the highest-impact path", or "what did the agent decide, and why did it stop?". Every answer cites the pipeline data it came from, and Thoth is strictly read-only — it can explain the scan, it cannot launch a new one or invent a finding.
Export the report
When you are ready to hand something off, download the PDF. It is a client-ready deliverable — findings with evidence, impact and remediation, plus the agent narrative — generated on demand from the assessment you just reviewed.
The screens behind the verdict
Beyond the assessment, four screens let you audit how the agent reached its conclusions — so the output is never a black box.
World-model
A visual map of hosts, services, subdomains, web endpoints and the edges between them — the structure the agent reasons over. Inspect a node to see what is known about it.
Attack chains
The multi-step paths where low findings combine into a high-impact outcome, each shown as a numbered kill-chain. An empty result is honest, not a failure — it means no genuine chain was found.
Loop decisions
The decide-loop trail: at each step, whether the agent continued or stopped, and the reasoning behind it. This is where you see the agent prove it bounds its own work.
Validation queue (admins)
Where active-payload probes wait for sign-off. The agent only proposes them — it never runs them itself. An admin reviews which finding and asset each one targets, then approves or denies; approval mints a one-time, scope-bound token before the probe fires.
- Re-scan as your surface drifts. The world-model compounds across scans, so repeated runs on the same scope get sharper, not noisier.
- Trust CONFIRMED, verify unproven. Treat confirmed findings as real and reachable; ask Thoth or run an approved probe before acting on the unproven ones.
- Use Thoth to interrogate, not to trust blindly. Every answer is grounded in the scan's data and cites it — ask it to show its work.