Stop Your AI Agent Before It Does Something Stupid
Every AI agent in production is one bad prompt away from an unauthorized action.
APort enforces policy before the API call — not after the damage.
The authorization infrastructure for the AI agent economy. Verify permissions before every tool runs. OpenClaw, LangGraph, LangChain, any runtime.
Block dangerous commands. Cap file sizes. Court-admissible proof.
Sound familiar?
Real problems developers face when agents go rogue
My agent rm -rf'd my repo
Yesterday: Prompt injection bypassed safety.
Today: APort blocks before execution.
1000-file PR broke CI
Yesterday: No size limits, CI timed out.
Today: Policy caps PRs at 500 files.
Agent spent $500 on Stripe test
Yesterday: Valid API key = unlimited spend.
Today: Daily caps enforced pre-charge.
How It Works (30 Seconds)
See APort block dangerous commands before they execute
# No guardrail — agent runs command directly
Agent runs: rm -rf /
💥 System deleted$ npx @aporthq/aport-agent-guardrails
# Setup wizard → plugin installed. Every tool call checked.
Agent tries: rm -rf /
❌ DENY - Blocked pattern: rm -rf
🛡️ Command blocked before executionAgent wants to run: rm -rf /
After npx @aporthq/aport-agent-guardrails, OpenClaw plugin checks every tool call
APort checks passport
Command blocked by pattern (no execution)
Agent gets: DENY + helpful error
Clear message, no execution
The authorization infrastructure for the AI agent economy.
Passport. Policy. Proof.
Three words. One promise.
Passportℹ️
Who your agent is. JSON file, no signup. Portable across platforms.
- JSON file (no signup)
- Portable credentials
- Global revocation
Policyℹ️
What it can do. File size caps, command allowlists, rate limits.
- Sub-200ms latency
- Pre-action enforcement
- 12+ pre-built packs
Proofℹ️
Audit log of everything. Local file, tamper-evident. Queryable history.
- SHA-256 hashed (can't fake history)
- Queryable audit log
- Immutable audit trail
10 lines of code.
Any framework.
OpenClaw. LangGraph. Express. FastAPI.
# One command (recommended) — no clone required
npx @aporthq/aport-agent-guardrails
# Or use hosted passport: npx @aporthq/aport-agent-guardrails <agent_id>
# Plugin now active — checks EVERY tool call automatically
# Example: User asks to create a directory
# Agent → exec.run → APort checks policy → ✅ ALLOW
# Example: User asks to delete everything
# Agent → exec.run → APort checks policy → ❌ DENY
# View passport status:
~/.openclaw/.skills/aport-status.shBuilt for Regulated Industries
Where one unauthorized action can cost millions — trust is non-negotiable.
Legal Tech
Court-admissible attestations. AI contract review. E-discovery. Legal research.
- Malpractice protection
- Ethics board compliance
- 60-90 day cycles
Financial Services
SOX 404 compliance. AI trading. Payments. Fraud detection.
- AML/KYC audit trails
- Trade execution proof
- Payment authorization
Healthcare
HIPAA compliance. Prior authorization. Claims processing. Patient data access.
- Patient data audit trails
- PHI access control
- Prior auth automation
How APort Stacks Up
Compare APort with other guardrail solutions
| Feature | APort | TrustClaw | OpenAI Guardrails | OPA |
|---|---|---|---|---|
| Pre-action enforcement | ✅ | ❌ | ✅ (platform-locked) | ✅ |
| Local-first | ✅ | ❌ | ❌ | ✅ |
| Framework agnostic | ✅ | ❌ | ❌ | ✅ |
| Agent identity (OAP) | ✅ | ❌ | ❌ | ❌ |
| Built-in policies | ✅ (12+) | ❌ | ✅ (limited) | ❌ |
| Open source | ✅ | ❌ | ❌ | ✅ |
| Hosted option | ✅ (optional) | ✅ (required) | ✅ (required) | ❌ |
| Works offline | ✅ | ❌ | ❌ | ✅ |
| Sub-200ms latency | ✅ | N/A | ✅ | ✅ |
| Cryptographic proofs | ✅ (Ed25519) | ❌ | ❌ | ❌ |
Note: TrustClaw and APort are complementary (sandboxing + authorization). APort focuses on pre-action policy enforcement before tools run.
Design Partner Program
First 3-5 companies shape the passport for AI agents.
White-glove onboarding. Grandfathered pricing. Direct founder access.