Dante is offering a free command-line scanner for loose API keys, open services, and risky AI-agent settings on personal computers and servers. Its pitch is straightforward: inspect what an agent could reach, show evidence for each finding, and keep the basic audit local.
The project appeared in October 29, 2026 coverage from Leverage AI News. Dante’s website provides no launch timestamp, so that date is a public-coverage reference, not a confirmed release date.
According to Dante’s product page, the scanner is MIT-licensed JavaScript, requires Node.js, and supports macOS, Windows, Linux, and SSH-accessible servers. Its default checks need no account or internet connection. Those capabilities and privacy assurances are developer claims; the available evidence contains no independent detection evaluation.
The useful distinction is scope. Dante audits exposure on demand or on an agreed schedule. It is a young configuration scanner, not a real-time malware engine.
What Dante Checks and How It Explains Findings
Dante focuses on resources that could become liabilities when an AI agent can read files and run commands with a user’s permissions. Its published description groups these into exposed secrets, open services, and risky system or agent settings.
That framing makes sense. A credential does not need to be inside a malicious program to be at risk. A legitimate agent with access to a project’s environment files may already have permission to read it. The security question becomes which sensitive resources the agent can reach and what could induce it to use them.
Dante says each verdict includes the evidence behind it. When the tool cannot read the necessary evidence, it reports an unknown result instead of assuming the check passed.
The website’s example illustrates why that distinction matters:
- A web-directory check reports no readable
.env, SQL dump, or Git-related files within the location searched. - An SSH check reports uncertainty when no service answers on port 22 and the relevance of the configuration cannot be determined.
A clean result for one directory is narrower than a clean bill of health for the machine. Likewise, an unanswered port does not explain every possible SSH configuration.
I’d put more weight on this evidence trail than on a reassuring verdict label. The useful output is the path, setting, or observation that lets someone understand and address a finding. An unknown result should remain on the review list, especially when it concerns credentials or remote access.
How to Inspect Dante Before Running It
Dante publishes a package-inspection workflow alongside its launch pitch. That is useful for a security utility that will examine sensitive parts of a computer.
The developer’s instructions start with Node.js installed and use npm to inspect and download the dante-watch package:
npm view dante-watch dependencies scripts
npm pack dante-watch && tar -xzf dante-watch-*.tgz
grep -rn 'execProgram.*"curl"' package/src
These are commands for readers to inspect and run themselves, not results reproduced for this article. They serve three different purposes:
- Check package metadata. The first command requests dependency and script declarations. The developer says it should show an empty dependency list and no package scripts.
- Download the published package. The second command retrieves an archive from npm and extracts it. The site directs readers to the JavaScript under
package/src. - Locate a documented network-call pattern. The third command searches that source for calls that invoke
curl.
The distinction between metadata and behavior matters. Having no declared dependencies reduces the number of third-party packages to inspect. Having no install scripts removes one automatic execution mechanism. Neither property proves that the program’s own code is harmless.
The grep command is also a starting point, not a complete network audit. It finds the specific code pattern the developer identifies; it cannot, by itself, prove that no other communication path exists.
The extraction and search commands use Unix-style tools. Windows users will need a shell that provides them or equivalent extraction and source-search tools. The commands are also unpinned, so recording the package version is sensible if you want a review that can be repeated later.
Start a Local Check
The documented launch command is:
npx dante-watch
At Dante’s prompt, the site tells users to enter:
check my computer
The initial npm download can require internet access. Dante’s offline claim applies to the installed scanner’s basic local checks, not to fetching the package.
For a cautious first run, I’d separate acquisition from inspection: download the package, review the code and metadata, then try the local check with networking disabled. That can help assess the offline behavior without granting optional network access. It would not measure how many security problems the scanner can detect.
Offline Checks Have Explicit Network Exceptions
Dante says its default mode sends no telemetry, performs no background update check, and needs no account. The same page identifies optional features that can contact external services after the user approves the displayed command:
| Feature | Published purpose | Practical distinction |
|---|---|---|
| Knowledge-pack fetch | Download additional checking rules | Updates require a separate network action |
| Shop catalog | Retrieve optional offerings | Separate from the basic local audit |
| Bait watch | Alert when a decoy credential is used | Network-connected monitoring, not an offline scan |
The developer describes knowledge packs as signed data containing rules, not executable code. A downloaded pack can be installed without a network connection, with its signature checked against a key built into the program.
That design separates acquiring new rules from running checks. The practical tradeoff is freshness: users choosing a disconnected workflow need a deliberate process for obtaining and reviewing updated packs.
The site also identifies a direct connection to test whether port 22 answers on a server the user names. Auditing an SSH-accessible server necessarily involves networking. “Offline” should therefore be read as a property of the basic local audit, not a promise that every Dante feature operates without connections.
Bait watch deserves separate consideration. The developer says it uses fake credentials as tripwires and sends an email if one is used. Its advertised price is . That is a different security mechanism, with different connectivity and pricing conditions, from reading local configuration files.
Frequently Asked Questions
4 questions
1Does Dante Work Without Internet Access?
The developer says the installed scanner’s basic local checks work offline without an account. Package downloads, optional features, and remote-server checks involve networking.
