The open-source, all-in-one security platform
Trusted by security teams at
41 target types across eight surfaces, mapped to security disciplines, examined by the full capability stack, and delivered as one audit-grade report — the whole platform, before you scroll.
Pencheff is the all-in-one security platform — built around three orthogonal commitments: methodology, coverage, and reporting. Each is enforced by the engine, not the operator.
Pencheff follows an adversarial methodology modelled on manual penetration testing — reconnaissance, authenticated coverage, business-logic probing, and exploit chaining — delivered with the consistency of automation.
Read the procedureInjection, access control, authentication, cryptography, client-side, infrastructure, cloud, and API — examined with Pencheff's first-party probes. Auxiliary tools are optional and operator-managed.
See the instrumentsEvery assessment yields a formal report with executive summary, letter grade, and evidence — mapped to OWASP Top 10, SOC 2, PCI-DSS, NIST 800-53, ISO 27001, and HIPAA categories.
Inspect a dossierEach engagement traces the same adversarial path — from passive reconnaissance to multi-step exploit chaining — so that two assessments of the same target, six months apart, are directly comparable.
Passive enumeration: subdomains, DNS, certificate transparency, public artefacts, technology fingerprint.
Authenticated and unauthenticated crawls, API discovery, endpoint inventory, parameter cataloguing.
Forty-nine instruments fired against the surface — injection, access control, OAuth, cloud, business logic.
Each finding re-fired with crafted payloads. Request and response evidence captured. False positives discarded.
Single findings composed into multi-step attacks: SSRF → metadata, XSS → session theft, IDOR → privilege escalation.
Forty-nine first-party instruments cover the modern attack surface — injection, access control, authentication, cryptography, client-side, infrastructure, cloud, and API — and can be composed with repo scanning, threat models, compliance rollups, SBOM output, and LLM red team workflows. Auxiliary tools (nmap, sqlmap, nuclei, ffuf, hydra, nikto) remain optional and operator-managed.
First-party probes — exhaustively catalogued, individually verified, weighted by severity. Re-examination of any finding is unlimited on every plan.
Web DAST, code analysis, dependency intelligence, infrastructure posture, AI red team, and cloud surface checks all normalise into the same finding schema — so an engineer triaging injection and an auditor reviewing compliance are reading the same record.
Pencheff covers the full OWASP LLM Top 10 (2025) with curated payload libraries, attacker-driven loops, and judge-backed scoring. Agentic testing probes tool calls, memory, planners, and swarm workflows, with provider-aware setup for LLM endpoints, speech endpoints, RAG, and agent memory. The Sentry guardrail enforces policy at runtime.
Roleplay, payload splitting, obfuscation, encoding variants, jailbreak corpora, and regression suites against any OpenAI-compatible endpoint. Attacker-LLM loops and judge-backed scoring when configured.
LLM red team →Tool authorization abuse, memory and context attacks, planner hijacking, cross-session leakage, retrieval poisoning, and swarm orchestration probes for agent-based products.
Agentic tests →Sentry runtime guardrail for prompt, response, tool, and PII policy checks. Maps to OWASP LLM, MITRE ATLAS, NIST AI RMF, EU AI Act, ISO/IEC 42001, and SOC 2.
Guardrails →Runtime protection · live traffic
Point your app at a target’s guardrail proxy and Pencheff inspects every request inline — prompts, responses, tool calls, and the memory your agents trust.
Drop-in gateway for any OpenAI-compatible model. Every prompt and response is checked against your OWASP-LLM policy before it reaches the model — or your users.
Runtime proxy →Per-target rules gate the tool calls your model makes — block SSRF, secret exfiltration, and destructive actions, require human approval, or redact credential-shaped arguments.
Agent firewall →Every proxied request recorded as a span tree — the LLM call, the detector verdict, the firewall decision — viewable per target, with zero added latency.
Runtime traces →Audit pasted rows, local memory files, Mem0, Zep, LangGraph Store, Redis, Pinecone, Chroma, Qdrant, Weaviate, and custom HTTP memory sources for secrets at rest and memory poisoning.
Memory scanner →From registration to remediation, the same procedure governs every assessment — so that the engineer who runs it, the operator who reviews it, and the auditor who reads it are all working from the same record.
Provide a target URL and, optionally, credentials for authenticated coverage. All secrets are encrypted at rest.
Commission a full Deep assessment. Every security scan runs Deep; progress streams live and stages are logged for review.
Triage findings with full request/response evidence. Re-examine any finding after remediation with a single action.
Download a formal DOCX or PDF report, dispatch to ticketing, and close out with verified evidence.
Two dossiers, issued together. Engineering receives the technical record; audit and executive readers receive the framework-mapped summary. Bound by a single grade and a single date of issue.
Pencheff normalises every finding against application security standards, audit frameworks, and AI governance requirements — so your engineers, auditors, and compliance officers work from the same report.
Every confirmed finding maps to one or more application security control frameworks.
Dossiers are issued with compliance rollup tables ready for auditor review.
AI assessment output maps to governance frameworks for LLM and agentic products.
Slack, Jira, GitHub, SARIF, webhooks, and more — findings route into your existing toolchain automatically. No new dashboard to babysit.
Dashboards, reports, multi-workspace, integrations, and schedules. Sign up and run a full assessment in under three minutes.
Get started →Native Mac client. Same workspace as the web app, plus on-device repo scanning, Downloads + macOS posture monitors, and an agentic remediation runner that keeps source on your machine.
Get started →Run deterministic checks in pipelines. Emit SARIF, trigger GitHub checks, and route findings into the platform from any CI system.
Get started →Connect Claude Code, Cursor, or Codex to mcp.pencheff.com and run the full engine from your editor — scan a repo, verify findings, and fix them with your IDE's own model. IDE-driven work consumes no Pencheff AI quota or credits, and results still sync to your dashboard.
Get started →Community Edition runs as Docker Compose under AGPL-3.0. Enterprise pulls private registry images; the org API key is checked before any image is fetched.
Get started →An LSP marks scan findings inline in VS Code, JetBrains, Neovim, and Emacs. The scan runs from the terminal, CI, or the platform; the editor highlights the line.
Get started →The scan is the centre. These are the surfaces that commission it, explain it, teach it, and carry the record into the tools you already run.
After the breaker swarm merges its findings, a commander sends specialist operatives back at the same result. An operative that finds a seam can spawn its own, down to a fixed depth, inside one shared session and one time budget.
One authorization covers every registered target in the organisation. The campaign fans those targets through the existing scan pipeline and streams progress as it moves from scope to report.
A free, hands-on ethical-hacking course at /academy. Lessons are written against real lab output, not invented screenshots, and sign-in is the same account as the rest of Pencheff.
Community Edition remains a Docker Compose stack under AGPL-3.0, with no licence fee. Enterprise images live in a private registry and are pulled only after the org API key is accepted.
Three opt-in mail flows. When mail is not configured, the rest of the platform keeps running and nothing is sent.
The same numbers as the findings register, drawn so a stakeholder can read them in a few seconds. Charts render only after a scan finishes.
Opt-in traces, logs, and metrics across the API, workers, the MCP plugin, external tool runs, and agent turns — plus a tamper-evident audit log. Off by default, so a stock install pays nothing for it.
A finding can be handed to a human pentest partner without leaving the record. The callback updates verification when their triage comes back.
Separate from the crawl. Traffic captured by the browser-extension proxy is turned into an OpenAPI 3.1 description of what the application actually served.
Every dependency advisory has its own page: the cached record, the NVD enrichment, and the priority signals used to decide whether it is urgent.
A language server publishes scan findings as inline diagnostics. The editor does not run the assessment; it shows the result on the line it belongs to.
The swarm’s Active Directory breaker runs beside the other specialists when the engagement includes that surface. Findings land in the same register as the web and cloud results.
Direct answers — on authorisation, scope, plans, audit acceptance, self-hosting, and credential handling. For anything else, write to us.
The Community Edition is free and open source under the GNU AGPL-3.0 licence. It is on GitHub and self-hostable as a Docker Compose stack, with no seat limits. Enterprise is a separate deployment: private registry images that pull only when the organisation API key is on an enterprise org.
Pencheff is for applications you own or have been granted written permission to assess. It is an instrument of assurance, not a means of unauthorised access. Please direct it only at systems within your mandate.
One complete engagement against a target: reconnaissance, infrastructure, injection, client-side, authentication, authorisation, advanced web, API, business logic, cloud, file handling, websocket, subdomain takeover, and exploit chaining. Re-examination of individual findings is unlimited.
Every security assessment runs the Deep profile: 30-90 minutes, contingent on application breadth. A separate load-test profile stays opt-in and is not a security scan.
Yes. DOCX and PDF reports include evidence-backed mapping to OWASP Top 10 (2021), PCI-DSS 4.0, NIST 800-53, SOC 2 (CC6/CC7), ISO 27001:2022, and HIPAA Security Rule - accepted by auditors as evidentiary material.
Yes. Community Edition is a Docker Compose stack under the GNU AGPL-3.0 licence. Enterprise pulls private registry images; the organisation API key is checked before any image is fetched, and a key that is not on an enterprise org cannot pull them.
Credentials are encrypted at rest with Fernet (AES-128 in CBC mode with HMAC-SHA256). Removing a target removes its credentials immediately.
A complimentary assessment takes under three minutes to commission and under thirty to complete. No credit card, no sales call.