Mission and limits
Objective, core business, surfaces, permitted actions and stop criteria.
Agents extend the investigation
Specialists preserve judgement
Mission-driven agents explore and correlate web and API journeys, preserving identity, state, hypotheses and evidence throughout the investigation. Every action happens within a defined scope and under a specialist's direction.
berghem-agent — zsh — 80×24
$ berghem-agent init --mode=autonomous> Mission: protect the critical asset defined in scope> Model: multi-model architecture (per task and sensitivity)> Targets: web, mobile, cloud, AD, network, API, containers, CI/CD> Status: Ready for autonomous security assessment> Initiating reconnaissance...
WHAT THE AGENTS DO
Mission, not scanning
A Berghem agent is neither a scanner nor a generic “web agent”. It receives a mission — the critical asset that must not be compromised — and investigates the routes that lead to it. Throughout the investigation the agent holds context: it preserves identities and state, compares profiles, forms and prioritises hypotheses, captures evidence and re-runs tests.
The specialist stays at the centre of the operation: defining the mission and the crown jewel, interpreting ambiguity, authorising sensitive actions and answering technically for the result. The agent extends reach; the specialist preserves judgement.
FLOW
Web → API → Web
A mission rarely lives on a single surface.
WebAPIWeb
A journey starts in the web application — a login, a session, a business rule — moves through an API call that exposes authorisation or a sensitive flow, and returns to the web layer to confirm the real impact on the user experience.
By crossing those surfaces while holding the same state and the same identity, the agent shows not only that a flaw exists, but that it can be chained all the way to the core business. It is this correlation across surfaces that separates an investigation from a checklist of isolated vulnerabilities.
This Is Not a Scanner
Berghem Agents don't just check boxes — they reason through attack chains like a senior penetration tester with 20 years of experience.
Reasons, Doesn't Pattern Match
Uses a multi-model architecture.
Works on Any Infrastructure
Web, mobile, cloud, Active Directory, network, API, containers, CI/CD — one agent covers your entire attack surface.
Active Hardening
Doesn't just find vulnerabilities — actively recommends and validates fixes, working standalone or fully integrated into your security pipeline.
Supported Infrastructure
Web ApplicationsMobile (Android/iOS)Cloud (AWS/Azure/GCP)Active DirectoryNetwork InfrastructureAPIs (REST/GraphQL)Containers & K8sCI/CD Pipelines
The Berghem Moat
20+ years
of penetration testing expertise for major financial institutions worldwide
10K projects delivered
of curated offensive security data powering our models
1 multi-model architecture
built via CPT, SFT, LoRA, and DPO on real-world pentest data
Capabilities
Offensive Testing
Autonomous penetration testing that adapts to your infrastructure and chains vulnerabilities like a human attacker.
Active Hardening
Goes beyond detection — validates remediation, tests fixes, and hardens configurations proactively.
Flexible Deployment
Deploy standalone, integrate with your SIEM, or embed in CI/CD pipelines. Works where you need it.
Continuous Assessment
Not a point-in-time test. Continuous monitoring and reassessment as your infrastructure evolves.
AGENCY
Automation runs steps
An agent pursues a goal within limits
The difference is not driving a browser. It is holding the mission, selecting tools, interpreting results and reframing hypotheses.
Objective
Observe
Hypothesis
Act
Verify
Correlate
Change route or stop
ARCHITECTURE
One architecture, different intelligence routes
Not every task needs the largest model, and not every piece of data can be processed in the same environment.
Intelligence
Specialised models, private models, authorised third parties or models inside your tenant.
Execution
Browser, APIs and specialised tooling; mobile within limited scopes.
Control and evidence
Identities, approvals, logs, circuit breaker, replay and reporting.
GOVERNANCE
Autonomy proportional to risk
Scope and access are defined before execution. Sensitive actions require approval and irreversible operations can be blocked.
Before
Mission, identities, domains, tooling, limits and evidence criteria.
During
Monitoring, parameter validation, checkpoints, cost ceiling and circuit breaker.
After
Replay, logs, evidence, retention, disposal, recommendations and retest.
Don’t just bring a URL
Bring the mission an attacker must not complete
Request a demo and see how autonomous security agents can transform your security posture.
Request Demo