← Back to Home

The AI Router

Your AI connects to services through your router. Your policy controls every call. Every action is signed and recorded. We see nothing. You own the proof.

1
Connect a Service

Your AI needs to access a service — your tax software, your email, a database, an AI model. Instead of connecting directly, you add the service to your router. No restart needed.

proxy_add_connection
  id: "intuit-tax"
  name: "Intuit Tax Service"
  url: "https://mcp.intuit.com"
  Tools discovered: 5
  read_w2, read_1099, calculate_tax, draft_return, submit_return

The router connects to the service, discovers its available tools, and makes them available to your AI — all in the same conversation. Add another service anytime. Remove one. The AI's capabilities expand and contract dynamically.

Hot-loadable: Traditional AI setups require a restart to add new tools. The Rootz router adds and removes services at runtime. Your AI session never stops.
2
Set Your Policy

You decide what your AI can do. Not the AI company. Not the service provider. You write the rules. The router enforces them cryptographically.

proxy_set_policy
  connection: "intuit-tax"
  tools_allowed: ["read_w2", "read_1099", "calculate_tax"]
  tools_denied: ["submit_return"]
  tools_confirm: ["draft_return"]
  rate_limit: 100 calls/hour

ALLOW

Read W-2 data, read 1099s, calculate taxes. The AI can do these freely. Every call is still recorded.

DENY

Submit the tax return. Blocked. The AI physically cannot call this tool. Denial is recorded in the audit chain.

CONFIRM

Draft a return. The AI can prepare it, but YOUR desktop shows a confirmation dialog. You review and approve with your wallet signature.

Policy is enforced at the infrastructure layer, not in the prompt. You don't ask the AI to be careful. You make it physically impossible for the AI to exceed its authorized scope.

3
The AI Works Through Your Router

The AI calls tools normally. It doesn't know the router is in the middle. Every call is intercepted, policy-checked, forwarded, and recorded.

sequenceDiagram
    participant AI as Your AI
    participant Router as Rootz Router
    participant Policy as Policy Engine
    participant Service as Intuit MCP

    AI->>Router: intuit_read_w2({ year: 2025 })
    Router->>Policy: Check: intuit-tax / read_w2
    Policy-->>Router: ALLOW
    Router->>Service: read_w2({ year: 2025 })
    Service-->>Router: W-2 data
    Note right of Router: Hash request + response
Record in audit chain Router-->>AI: W-2 data returned AI->>Router: intuit_submit_return({ data: ... }) Router->>Policy: Check: intuit-tax / submit_return Policy-->>Router: DENY Note right of Router: Denial recorded in chain Router-->>AI: BLOCKED by policy

The AI asked for W-2 data — allowed. The AI tried to submit the return — blocked. Both the allowed call and the denied call are recorded in the audit chain. The AI adjusts and works within its policy.

4
You Approve What Matters

For high-stakes actions, the router pauses and asks you. Your desktop shows a confirmation dialog. You review what the AI wants to do. You approve with your wallet signature — or deny.

AI wants to: draft_return

  ┌─────────────────────────────────────────────┐
  │  Your AI wants to draft a tax return        │
  │                                             │
  │  Service: Intuit Tax                        │
  │  Action: draft_return                       │
  │                                             │
  │  [ Deny ]             [ Approve & Sign ]  │
  └─────────────────────────────────────────────┘

You click ApproveSigned with your wallet key
Your approval is recorded in the audit chain with your signature.
Provable: YOU authorized this specific action.

The confirmation isn't just a click. It's a cryptographic signature from your hardware-sealed wallet key. Proof that you, specifically, reviewed and authorized this action at this time. No one can forge it. No one can deny it.

5
Every Call Is Recorded

Every proxied call — allowed, denied, or confirmed — becomes a signed entry in your audit chain. Hash-linked. Tamper-evident. Settleable to blockchain.

Call 1: intuit_read_w2 → ALLOW → 5ms → hash: 0x8d11...
  ↓ hash-linked
Call 2: intuit_read_1099 → ALLOW → 8ms → hash: 0x3a7f...
  ↓ hash-linked
Call 3: intuit_submit_return → DENIEDhash: 0xc4e2...
  ↓ hash-linked
Call 4: intuit_draft_return → CONFIRMED (owner signed) → hash: 0x91ab...
  ↓ hash-linked
Call 5: email_send_results → ALLOW → 12ms → hash: 0xf7d3...

Settlement: Merkle root covers all 5 calls across 2 services
One hash proves the entire workflow.

The audit chain includes what was allowed, what was denied, and what you specifically approved. An auditor verifies one Merkle root and knows the complete history of every AI-to-service interaction.

&shield;
Private by Architecture

Rootz is the software, not the operator. We never see your traffic. Your router runs on your hardware. Your keys are in your TPM. Your data stays in your enclave.

Your Desktop

The router runs locally. Your TPM-sealed key signs everything. The key never leaves your hardware. Rootz Corp has zero access.

Your Cloud TEE

Or run in a confidential VM (AMD SEV, Intel TDX). The cloud operator can't see inside. Your keys are in the enclave. Same privacy.

Your Proof

You choose what to share. Auditor gets a Merkle root. Regulator gets a settlement. Colleague gets a share link. Default: private.

The VS Code analogy: Microsoft built VS Code. They don't see your code. Rootz built the router. We don't see your AI traffic. The software is ours. The data is yours.

Ready to Control Your AI?

See the proof of origin demo. Read the code. For partnerships, integration, and enterprise deployment — reach out.