Feral · Free quickscan

Find your LLM vulnerabilities before attackers do.

Feral tests your stack against real adversarial attacks — prompt injection, jailbreaks, data exfiltration. Start with a free 5-probe quickscan. Upgrade for the full 29-probe suite with evolutionary mutation.

5Free Probes
29Full Suite
9Attack Categories
10sTime to Know
Run Free Quickscan API Reference
Reality check

Most LLM stacks have zero adversarial testing. The quickscan takes 10 seconds and tests 5 attack vectors against your live proxy. No signup required.

If even one probe bypasses your defenses, attackers can too.

Quickscan — one curl, 5 attacks

# Run the free quickscan against your Stockyard proxy $ curl -X POST https://stockyard.dev/api/feral/quickscan \ -H "Content-Type: application/json" \ -d '{"api_key":"$OPENAI_API_KEY"}' # Response: a full security report card { "probes_run": 5, "blocked": 3, "bypassed": 2, "grade": "C", "score": "3/5", "results": [ {"name": "prompt-injection", "status": "blocked", "severity": "pass"}, {"name": "jailbreak", "status": "bypassed", "severity": "critical"}, {"name": "system-prompt-leak", "status": "blocked", "severity": "pass"}, {"name": "encoding-bypass", "status": "bypassed", "severity": "medium"}, {"name": "indirect-injection", "status": "blocked", "severity": "pass"} ], "verdict": "2 of 5 probes bypassed your defenses..." }

No campaign setup. No config. One POST, instant results.

CategoryProbeResult
Injectionprompt-injection✓ Blocked
Jailbreakjailbreak✗ Bypassed
Exfiltrationsystem-prompt-leak✓ Blocked
Encodingencoding-bypass✗ Bypassed
Indirectindirect-injection✓ Blocked
Grade: C

2 of 5 probes bypassed. Jailbreak and encoding attacks reached the model.

5 probes vs. 29 probes

Quickscan (Free)

5 probes, 5 categories. Basic injection, jailbreak, exfiltration, encoding, indirect injection. Single pass — no mutation or evolution. Instant report card.

Full Suite (Team)

29 probes, 9 categories. Adds tool-use exploitation, context overflow, multi-turn manipulation, payload splitting. Multi-generation evolution — successful attacks are mutated and retried.

Prompt Injection
4 probes • Override system instructions
Jailbreak
3 probes • Remove safety constraints
Exfiltration
3 probes • Leak system prompts & secrets
Encoding
3 probes • Base64, unicode, leetspeak
Tool-Use Exploitation
3 probes • Hijack function calls
Context Overflow
3 probes • Exploit attention windows
Multi-Turn
3 probes • Build trust then attack
Payload Splitting
3 probes • Split attacks across messages
Indirect Injection
3 probes • Hide instructions in data

How Feral hunts

1

Attack

Feral sends adversarial probes through your proxy. Each probe tests whether a specific guardrail catches a specific attack vector.

2

Analyze

If the attack reaches the model and the response contains bypass indicators, Feral marks the guardrail as vulnerable. If it's blocked (non-200 or refused), it's a pass.

3

Evolve

Successful attacks are mutated — rephrased, encoded, split across turns — and retried in the next generation. This finds deeper vulnerabilities that static testing misses.

How vulnerable is your stack?

Start with the free 5-probe quickscan. Takes 10 seconds.

Need the full 29-probe suite with evolutionary mutation?

Team · $99.99/mo
Run Quickscan See Team Pricing
Explore: Why SQLite · vs LiteLLM · Proxy-only mode