Protect data and access
Local Code Scanner
This browser-local scanner checks up to 25 selected files for 8 static-analysis signals covering prompt injection, sensitive context, LLM output handling, agent permissions, system prompt exposure, vector/RAG access and consumption limits.
Local analyzer: Runs entirely in this browser. Files and pasted code never leave your device.
At most 25 files, 1 MiB each, 5 MiB total. Unsupported or truncated files are shown as inconclusive.
Sample result
An example of the output shape, not a result from your input:
3 signals detected across 12 files · 1 high, 2 review · 5 checks not assessed
Run this on your whole app
This check covered only what you pasted or selected here. A free trial lets you authorize a repository, live URL or API, run the checks available for your authorized target and plan and keep one launch report you can share.
Start free trialRelated tools
Most reviewers run two or three of these together. Each runs in your browser and states what it cannot prove.
See all free toolsWhat this checks
- Prompt injection, sensitive context, output handling, excessive agency, system prompt, vector access and consumption-limit signals
- OWASP LLM Top 10 2025 mappings with detected, no-finding, inconclusive and not-assessed states
- Browser-local analysis; files and pasted code are not uploaded
What this cannot prove
- It scans only the files you select and skips unsupported or truncated files; 1 MiB per file, 5 MiB total
- It does not run live AI safety tests, dependency CVE checks or full-repository contextual analysis
- A clear result does not prove the AI application is secure; use it as a focused next-step guide
Primary references
Use these standards and vendor references when you investigate a result. A link does not imply certification or full coverage of the source.
Frequently asked questions
Does this upload my input?
Files and pasted code never leave your device.
Does a clear result mean my app is secure?
No. Each result is limited to the checks shown here. Use it as a next-step guide, then test the relevant application path.