Generates Python Unit Tests and Security Probes for Your Code — Results Are Proved by Running in a Sandbox.
Most LLM agents generate plausible-looking tests and stop. Covergent runs each generated test in a sandbox and reports what actually passed, how much coverage it added, and which vulnerabilities it caught. The results are the deliverable, not the text.
No account needed to run it. Sign up to save your run history.
| Target | Unit tests | Probes | Vulnerabilities caught | Coverage |
|---|---|---|---|---|
find_user (SQL injection) | 7/9 pass | 3 | 2–3 ⚠ | ~80% |
find_user_safe (fixed) | 9/9 pass | 3 | 0 ✓ | ~75% |
Same generator, same probes — the numbers move because the code changed. Every number is produced by real execution.
Parses your repo by function/class boundaries and keys a CWE/OWASP knowledge base off what each function actually does.
DeepSeek writes unit tests (style-matched to your repo) and adversarial security probes grounded in the retrieved patterns.
Runs everything in a sandbox — pass rate, coverage delta, vulnerabilities caught. A failing probe is a real bug found.