---
name: privi-callproof
description: Use Privi CallProof to authorize, bound, approve, and cryptographically record AI-agent tool-call intents for verified Privi Pro customers. Never treat a signed executor report as proof of external execution.
---

# Privi CallProof — agent trust plane

Product: https://privi-ai-21-gateway.onrender.com/callproof

Pro console: https://privi-ai-21-gateway.onrender.com/callproof/pro

MCP endpoint: https://privi-ai-21-gateway.onrender.com/mcp

Pricing: Included with Privi Pro Unlimited, $35 one-time covering all Privi services. This skill is not a separate purchase or financial payment rail.

Activation: verified Stripe Privi Pro wildcard entitlement, one-time operator-issued Pro token, durable PostgreSQL and Ed25519 signing key. Self-service credential issuance is not yet enabled. A Stripe receipt or customer ID alone is not an API key.

## Threat model

The requesting agent cannot approve its own write-class request with its agent credential, and it cannot report the separately credentialed executor's outcome. The Pro owner issues short-lived capabilities with exact tool allowlists, bounded calls, and estimated policy budgets. The executor must enforce its own authenticated authorization and idempotency. Active Pro entitlements are rechecked at use time, including after refunds/revocations. CallProof itself never executes third-party tools.

Do not put the owner token, executor credential, signing key, customer email, or private application keys into an agent prompt, public tool argument, repository, or logs. Credentials belong in private integration headers and a secure vault.

## Agent / executor workflow

1. The verified Pro owner creates a grant via POST /v1/callproof/grants using its Pro bearer credential. Supply tenant label, actor ID, short TTL, allowed tool names and read/write effects, max calls and an estimated-policy budget in cents. This is a safety quota, not a monetary charge.
2. Receive a one-time agent capability token (prefix privi_cap_) and a separate executor reporting token (prefix privi_exec_). Store them in separate secret vaults, never send both to an AI model.
3. Hash the intended input locally using SHA-256. The agent reserves one tool intent via POST /v1/callproof/intents using its restricted capability credential. Example JSON:
   {"idempotency_key":"example-operation-0001","tool":"crm.update","input_sha256":"aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"}
   An identical retry returns the same intent; a changed payload using the same key is rejected.
4. For write-class tools, a separate owner with the Pro credential approves the intent through POST /v1/callproof/approve. Never delegate owner approval token to an autonomous agent.
5. The executor checks that the intent is authorized, performs the action independently through its own authenticated external API using the original idempotency key, then reports the actual outcome via POST /v1/callproof/report using its executor reporting credential.
6. Privi stores a signed receipt of the executor-reported outcome containing actor, tenant, request/response hashes, UTC timestamp and previous receipt hash. The Pro owner can verify through POST /v1/callproof/verify and export the last 30 signed receipts from GET /v1/callproof/receipts.
7. Revoke a leaked grant through POST /v1/callproof/revoke. Remove customer Pro entitlements on refunds/disputes to block further access.

## MCP tool entry points

Two additional tools appear in Privi's existing remote MCP catalog, and both require separate valid credentials:

- privi_callproof_reserve: delegated agent bearer, scoped retry-safe policy intent reservation. Does not execute the requested external tool.
- privi_callproof_verify: verified Pro owner bearer, verifies signed executor-reported receipt and prior hash link.

Pro activation, owner approval and executor reporting are deliberately NOT general-purpose agent tools; they require independent authenticated HTTPS APIs.

## Evidence limitations

An Ed25519 signature shows that Privi signed an executor's report. It does NOT independently establish an external action happened, guarantee exactly-once execution at a third-party API, prove funds moved, provide OAuth/OIDC, or create blockchain finality. The stored hash chain is not a public immutable ledger. There is no automatic PRVI financial mainnet anchoring or security certification.

Customer payment alone does not automatically provision access; production activation requires the database, signer and operator credential to pass their gates.

Copyright (c) 2026 Privi World Corp. All rights reserved.
