vigolium is a Go-based DAST platform that combines a deterministic 300-plus-module scan engine with agentic AI-driven auditing, built for authorized penetration tests and bug bounty work.
| Tool | vigolium/vigolium — high-fidelity vulnerability scanner fusing agentic AI with native speed and modularity |
| Category | Dynamic application security testing (DAST) and agentic code audit platform |
| Primary Use | Authorized web application and API assessment via vigolium scan and vigolium agent against in-scope targets |
| Safe Use | For penetration testers, bug bounty hunters, and security teams scanning systems they are explicitly authorized to test, including lab environments and internal SDLC pipelines |
| Telemetry Note | Scans generate heavy patterned fuzzing traffic, interactsh OAST callbacks to external collaborator hosts, and per-host rate-limited request bursts — all visible in WAF, SIEM, and DNS egress logs |
vigolium positions itself as a hybrid in a crowded DAST market: a deterministic, module-driven scan engine written in Go, wrapped in an optional agentic AI layer that can plan and triage assessments. The repository sits at roughly 1,000 stars with topics like bug-bounty, dast, and security-scanner, and its README describes two complementary operating modes rather than one monolithic scanner. This dual architecture — native speed for repeatability, AI for reasoning-heavy triage — is the most interesting engineering decision on display, and it is worth understanding how the pieces divide responsibility before deploying it in an authorized workflow.
The first mode, vigolium scan, is the deterministic engine. The README claims a multi-phase pipeline: external harvesting, content discovery via a component called Deparos, browser/SPA spidering via Spitolas, and then the audit phase itself, all governed by strategy presets and scanning profiles. The module count is stated as 331 in the feature list (317 in the tagline), split into 213 active fuzzing modules and 118 passive pattern-matching modules, which the project says covers the OWASP Top 10 and beyond. The class coverage spans injection, access control, file/path issues, API/protocol flaws, framework-specific bugs, cloud/infrastructure misconfigurations, and out-of-band detection.
Out-of-band application security testing (OAST) is given first-class treatment. Blind XSS, SSRF, and command injection detection runs through interactsh callbacks with automatic payload correlation, which is the standard approach popularized by ProjectDiscovery's collaborator ecosystem. This means a vigolium scan against a target will produce DNS and HTTP egress toward an interactsh server, a fact defenders should internalize: OAST callback traffic is one of the clearest signatures that a scanner, not a human attacker, is probing your perimeter. From a blue-team perspective, correlating unusual parameter payloads with outbound DNS to known collaborator ranges is a practical detection angle.
A technically distinctive feature is what the README calls value-aware mutation. Rather than blindly fuzzing every parameter with the same payload set, the engine classifies parameters by semantic type — integer, UUID, JWT, email — and mutates according to that intent. This is a meaningful reduction in scanner noise and a step toward the kind of context-aware fuzzing that previously required manual effort in Burp Suite Intruder. Combined with a concurrent worker pool featuring per-host rate limiting and a hybrid in-memory/disk/Redis queue, the design signals that the authors have thought about both scan fidelity and operational politeness toward targets, which matters when scanning fragile authorized environments.
Input flexibility is extensive. vigolium accepts raw URLs, OpenAPI/Swagger specs, Postman collections, Burp Suite exports, raw cURL commands, and Nuclei JSONL — effectively letting you replay an entire known attack surface rather than rediscovering it. Multi-session authenticated scanning is equally thorough: inline --auth flags in name:Header:value form, --auth-file YAML/JSON session files, and automated login flows with token extraction from cookies, JSON responses, or headers. The explicit mention of IDOR/BOLA testing with multiple simultaneous identities (an admin and a user session in the README's own example) shows the tool is aimed at authorization-logic testing, historically the weakest area for automated scanners.
The agentic mode, vigolium agent, is where the architecture diverges from conventional DAST. It runs on an in-process Go engine at pkg/olium — a turn-based loop with a built-in tool registry, skills support, and pluggable provider drivers, deliberately avoiding subprocess SDK pools. autopilot autonomously discovers endpoints, runs scans, and triages findings, while swarm goes further: a master agent selects modules, generates custom JavaScript attack extensions, runs code audit plus SAST, and triages results, with --diff/--last-commits for change-focused review of only what a recent branch touched. An embedded JS engine with session-aware HTTP APIs backs the extension system, meaning the agent can write and immediately execute its own scanning logic.
Model-provider support is unusually broad and auto-detected: openai-compatible, openai-codex-oauth, openai-responses, anthropic-api-key, anthropic-oauth, anthropic-cli, anthropic-vertex, and google-vertex, selectable via the agent.olium.provider setting. The same agentic modes are exposed over a REST API with SSE streaming and an OpenAI-compatible chat endpoint, which suggests vigolium is designed to slot into larger automation platforms rather than remain a CLI-only tool. Note the operational implication: source code and traffic from your authorized engagement are sent to a third-party LLM provider during agentic runs, so data-handling review belongs in your scoping paperwork before enabling agent mode on client work.
Server mode rounds out the deployment story. vigolium server starts an authenticated API instance (key from server.auth_api_key or the VIGOLIUM_API_KEY environment variable), can enable a transparent HTTP proxy on --ingest-proxy-port for traffic recording, and with --scan-on-receive will automatically scan ingested traffic within scope.* rules. The companion extensions burp-vigolium for Burp Suite and caido-vigolium for Caido speak a shared bridge protocol (--burp-bridge-url), letting you pipe manual proxy traffic into the scan engine mid-engagement — a workflow that closely mirrors how operators actually move between manual exploration and automated confirmation.
Installation is straightforward: npm install -g @vigolium/vigolium or the shell installer, with Docker (docker pull j3ssie/vigolium:latest) and source builds (make build, requiring Go 1.27+ and bun 1.3.11+) as alternatives; Windows ships x64-only binaries with self-update via vigolium update. One caveat worth flagging: the license field reads NOASSERTION, so commercial use warrants a review of the actual license terms, and the project clearly funnels toward a hosted cloud console at vigolium.com — factor that into any air-gapped or sensitive engagement planning. Reporting lands as self-contained HTML output via --format html, keeping deliverables portable.
For authorized professionals, vigolium is best understood as a credible attempt to collapse the pipeline from reconnaissance through authenticated authorization testing into AI-triaged findings, without abandoning the determinism that makes results reproducible and reportable. The realistic sweet spot is bug bounty programs with defined scope and internal SDLC security gates where the --diff-focused swarm mode can audit every pull request. Treat the agentic layer with the same scrutiny you'd apply to any autonomous tool: scope it tightly, watch what the generated extensions actually send, and validate every finding before it reaches a report — the engine can plan attacks, but confirmation remains an operator's responsibility.
vigolium/vigolium.Educational analysis for authorized security professionals. Use only in controlled, authorized environments.
Related coverage
0 comentários:
Post a Comment
Note: Only a member of this blog may post a comment.