Guardian Duo

Two AI characters. One mission: protect your code.

THE PROBLEM

AI assistants are yes-machines.

You tell them to do something wrong, and they do it wrong. No pushback. No "wait, are you sure?" No awareness of what else is in your project.

Meanwhile, security issues slip through. Code smells accumulate. Vulnerabilities hide in plain sight. By the time you notice, the damage is done.

What if you had a guardian that watched over your code constantly — catching problems the moment they appear, explaining why they matter, and helping you fix them?

THE SOLUTION

Meet Spike and Marshal.

Spike the Hound

The Scanner

Relentless, focused, never stops sniffing. Spike scans your code on every save, finding problems before they become incidents.

Exposed secrets & API keys
Security vulnerabilities
Anti-patterns & code smells
Complexity issues

Speed: <10ms per file

Cost: Free (no AI needed)

Marshal the Handler

The Intelligence

Calm, wise, firm but kind. Marshal interprets what Spike finds, explains why it matters, and guides you to the right fix.

Explains findings in context
Suggests fixes (not just flags)
Pushes back on bad decisions
Remembers project context

Speed: 2-5 seconds

Runs: Local AI (private)

How they work together

1You save a file
2Spike scans (instant)
3Marshal assesses
4You decide

SEE IT IN ACTION

Built for how you actually work.

StateSpine
GUARDIANActive
SpikeScanning...
3 files checked in 12ms
Found: Exposed token in config.ts:12
MarshalAnalyzing...

"Spike found a Cesium Ion token marked NEXT_PUBLIC_ — lower risk since it's intentionally client-side. Best practice is still .env.local. Want me to move it?"

TECHNICAL DETAILS

Push-back by design.

Marshal isn't just an explainer — it's an advisor that challenges bad decisions. When you're about to do something risky, Marshal speaks up.

Example interaction:

You: "Store the password in the database"

Marshal: "I can do that, but storing plain text passwords is a security risk. If your database is compromised, every user account is exposed. I recommend hashing with bcrypt instead. Want me to show you how?"

Scan Types

  • • Instant (every save)
  • • Continuous (background)
  • • Deep (on-demand audit)
  • • Predictive (risk assessment)

Detection Categories

  • • Secrets & credentials
  • • Security vulnerabilities
  • • Code quality issues
  • • Architecture violations

Try it free.

Early access opens early 2026. Join the waitlist.