Skip to main content

The Agent Firewall for High-Stakes AI

Deploy autonomous AI agents in legal, finance, and compliance — with the guardrails, approvals, and audit trails that regulators demand.

Try Free (5 min setup)Book Pilot ($15K-$25K)
✓ Self-hosted (your infrastructure) • ✓ Apache 2.0 open source • ✓ Production-ready in 2-4 weeks

What You Get

🛡️

Policy Enforcement

ALLOW, DENY, or ESCALATE decisions based on manifests, budgets, and risk hooks. Non-bypassable enforcement at the action boundary.

Learn more →
👤

Human Approvals

High-risk actions escalate to operators for review. Web UI + API with 5-minute SLA. Full audit trail of every decision.

Learn more →
📜

Tamper-Evident Logs

Hash-chained, Ed25519-signed interaction records. Cryptographically verifiable audit logs for regulators and courts.

Learn more →

What Customers Say

"The pilot paid for itself in the first month."
— Managing Partner, IP Litigation Boutique
"Got to production in 3 weeks. The audit trail was exactly what regulators wanted."
— CTO, Series B Fintech
"We can finally say 'yes' to AI agents without sacrificing governance."
— Director of Compliance, Regional Bank

See It In Action

Every agent action flows through the gateway for policy enforcement and audit logging.

# Agent proposes action
curl -X POST /gateway/execute \
  -H "X-API-Key: $KEY" \
  -d '{
    "uapk_id": "settlement-bot",
    "action": {
      "type": "legal",
      "tool": "send_settlement_offer",
      "params": {"amount": 5000}
    }
  }'

# Gateway response
{
  "decision": "ALLOW",
  "executed": true,
  "interaction_id": "int-abc123"
}

What Just Happened?

  • ✓ Manifest check: Is settlement-bot registered?
  • ✓ Capability check: Can it send settlements?
  • ✓ Budget check: $5K under $50K threshold
  • ✓ Policy check: Passed all rules
  • → Executed via connector
  • 📝 Logged with hash chain + signature

Before vs After UAPK Gateway

❌ Without UAPK

  • Compliance blocks every agent deployment
  • "Who authorized this?" (no attribution)
  • "Can we prove it in court?" (no audit trail)
  • "How do we stop it?" (no kill switch)
  • Months of back-and-forth with legal/compliance
  • Vendor logs (90-day retention, not court-admissible)

✓ With UAPK

  • Policy enforcement: ALLOW/DENY/ESCALATE
  • Attribution: Every action traced to agent + manifest
  • Court-ready logs: Hash-chained, Ed25519 signed
  • Human approvals: High-risk actions reviewed
  • Production in 2-4 weeks: Fixed-fee pilot
  • Your evidence: Self-hosted, indefinite retention

The "47ers" Library

Pre-built governance templates for common workflows

⚖️ Legal

  • IP Settlement Gate: Auto-negotiate up to $50K, escalate above
  • DMCA Takedown: 200 notices/day with compliance tracking
View templates →

💰 Finance

  • Trading Gate: $10K auto-execute, $100K daily cap
  • KYC Onboarding: Risk-based routing + sanctions screening
View templates →

🔒 Compliance

  • Vendor Due Diligence: Automated risk assessment
  • Email Guard: Rate limits + recipient validation
View templates →

Why UAPK Won't Be Obsolete

Model vendors will improve. Your governance requirements won't change.

🎯 Model-Agnostic by Design

UAPK governs actions at the boundary to real systems. It doesn't care which model you use (GPT-4, Claude, Llama, Gemini).

🏛️ Regulation Requires It

SOC2, GDPR, SEC audits require organization-owned evidence. "Check the OpenAI logs" doesn't work in court.

Ready to Ship Agents Safely?

Choose your path to production

🔬 Self-Host (Free)

Deploy in 5 minutes with Docker Compose

Apache 2.0 • Full source code • Community support

Start Now

🚀 Pilot Program

Production-ready in 2-4 weeks with expert help

$15K-$25K fixed fee • Manifests + training + 30-day support

Book Pilot