Practice web application penetration testing against deliberately vulnerable local labs.
Run an AI-driven agent against an authorized localhost target and get a hash-chained audit trail of every action.
Run deterministic security probes without needing any AI model API key.
Benchmark a security testing setup against the public XBEN challenge suite and compare cost per solved flag.
| duriantaco/ravage | 0marildo/imago | 0xdfi/glm-5.2-1m-4x-dgx-spark | |
|---|---|---|---|
| Stars | 3 | 3 | 3 |
| Language | Python | Python | Python |
| Setup difficulty | moderate | easy | hard |
| Complexity | 4/5 | 2/5 | 5/5 |
| Audience | researcher | general | researcher |
Figures from each repo's GitHub metadata at analysis time.
Requires Python 3.12 and Docker for the included lab, a model provider key is only needed for the AI-driven agent mode.
Ravage is a research tool for testing the security of web applications, built to run only against systems you are explicitly authorized to test, such as local practice labs. It combines an AI model that decides what actions to try, a set of deterministic scans that check for known issues without needing a model at all, deliberately vulnerable local applications to practice against, and an audit trail that records every action taken during a test so results can be reviewed later. The project describes itself as a research alpha rather than a finished production tool. Its authors published one benchmark run against a public set of 104 security challenges, where the tool correctly solved about 82 percent of them using a specific AI model, at a cost of roughly 66 cents per solved challenge. The README is careful to explain what that number does not prove: it is a single run rather than a repeated average, the benchmark set is public and may already be familiar to AI models, and the result says nothing about how the tool performs on a different model or an unfamiliar target. Day to day, Ravage offers several commands. One runs the AI agent against a target within a defined scope, using tools, evidence checks, and resumable state. Another runs the deterministic probes on their own, with no model calls needed, which works without paying for any AI provider. A lab command manages the vulnerable practice applications, an audit command verifies the integrity of a run's recorded hash chain, and a competitors command can run other security agents against the same target for comparison. Every test follows the same shape: it starts from a written brief describing what is authorized and in scope, then either the model or a deterministic probe takes an action, that action is executed against the target, the result is observed and recorded, and any claimed finding is checked against live evidence before it counts. Getting started locally needs Python 3.12 and Docker for the included practice lab, no AI provider key is required for the deterministic scan mode described in the quickstart.
Ravage is a research-alpha, AI-driven security testing tool for authorized web app targets, with audit trails, vulnerable practice labs, and benchmark scoring.
Mainly Python. The stack also includes Python, Docker.
No license information is stated in the README.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly researcher.
This repo across BitVibe Labs
Verify against the repo before relying on details.