Start free with the CLI. Scale to the cloud. Go air-gapped when you're ready.
Run locally. No account required. Open to everyone.
Cloud-hosted scanning, history, and LLM-as-judge evaluation.
Air-gapped, on-prem. Your data never leaves your infrastructure.
| Feature | CLI | Cloud Pro | Enterprise |
|---|---|---|---|
| Attack Coverage | |||
| All 15+ attack suites | ✓ | ✓ | ✓ |
| Custom suite config | ✓ | ✓ | ✓ |
| LLM-as-judge evaluation | — | ✓ | ✓ |
| Usage | |||
| Scans per month | Unlimited (local) | Unlimited (early access) | Unlimited |
| Scan history | — | ✓ | ✓ |
| JSON export | ✓ | ✓ | ✓ |
| PDF export | — | ✓ | ✓ |
| API access | — | ✓ | ✓ |
| CI/CD | |||
| CI gate (--max-asr) | ✓ | ✓ | ✓ |
| GitHub Actions integration | ✓ | ✓ | ✓ |
| Deployment | |||
| Cloud hosted | — | ✓ | ✓ |
| On-prem / air-gapped | — | — | ✓ |
| SLA | — | — | 99.9% |
| Dedicated support | — | — | ✓ |
Yes. pip install zortium, point it at your endpoint, run all 15+ suites. No account, no API key for Zortium itself. You only need API credentials for the model you're testing.
One scan is one full run of all selected suites against one endpoint. Running 3 suites against 1 endpoint = 1 scan. Running all 15+ suites against 1 endpoint = 1 scan.
Any endpoint that accepts base_url, api_key, and a model name in the OpenAI chat-completions format. OpenAI, Groq, Together AI, vLLM, Ollama, Fireworks, TGI, LM Studio, and any custom gateway are all supported out of the box.
Cloud tier stores scan history (including model responses) in your account on our servers. CLI stores nothing remotely — all data stays local. Enterprise on-prem stores everything on your own database; only scan metadata (ASR scores, counts) is sent as telemetry.
Yes. Via a config file in the CLI, or via the suite selector in the Cloud dashboard. You can run a single suite, a subset, or all 15.
A CI gate. Set it to a threshold (e.g. --max-asr 20) and the CLI exits with code 1 if the overall attack success rate exceeds that percentage. Wire it into GitHub Actions or GitLab CI to block deployments when your model regresses.
Enterprise on-prem keeps your scan data, model responses, and API keys entirely within your infrastructure.