One run, start to finish: a brief becomes a plan, recon maps the target, the agent proposes candidates, and the validator throws out everything it cannot prove
Want to see it work? Request a Demo, or head to gimorra.vigolium.com.
Safety by construction, not by babysitting
Most offensive agents ask a human to approve every action. That is not caution — it is an admission that the agent could do real damage if nobody is watching. Gimorra closes that gap in code. Three controls make a run incapable of hurting the target, so supervision becomes a dial you set once instead of a seat you sit in.The validator decides
The agent proposes. Code disposes. Every finding is re-proven by deterministic, non-AI code using the least invasive action that settles it. Nothing self-certifies.
Cleanup and a kill switch
Whatever a run touches, it puts back. One command ends the engagement: queued work goes unclaimable, and no further proof is accepted.
Scope guard on every call
Deny wins. Ambiguity blocks. Checked on every single tool call, before anything else gets a vote.
Point it at anything
One dial sets how hard it pushes, from fully passive to aggressive. Nothing else needs tuning to get a first result.
What lands on your desk
Bugs, not maybes
Every finding in the report was reproduced by deterministic code. No triage queue of plausible-looking noise.
Receipts for all of it
A hash-chained evidence trail ties each finding to the exact exchange that proved it.
A report you can forward
Verified-only Markdown or HTML, plus a live dashboard while the run is still going.
Memory between runs
Surface, tradecraft and accounts carry forward, so the second engagement starts where the first stopped.
Where it sits
Gimorra runs its tools on the machine you start it on, so that machine is the blast radius. Start it in a container — there is an all-in-one image for exactly this.
Curious what it finds on your own target? Request a Demo, or email [email protected].
