P402.io is an attribution and metering platform for AI usage costs, with the core goal of making AI spend accountable. It turns every AI request into an economic event, recording the event's owner, workflow, model, provider, token count, cost, budget, policy outcome, output result, and evidence status. The platform is privacy-focused by default, uses metadata-only metering, requires no storage of prompts, and offers a metadata-only mode from the very first event.
In terms of use, P402 mainly solves the problem that AI spend is discovered only after the fact and is difficult to attribute. Typically, provider invoices show only the total after usage, without indicating which customer, feature, workflow, department, employee, model, or retry caused the cost. P402 records attribution information at the moment an event occurs, helping organizations understand the gross margin of AI features, track AI cost of sales, feature gross margin, customer cost, retry waste, context waste, and cost per accepted output, and view the spend distribution by department, employee, workflow, model, and provider.
Main capabilities include: through the embedded MCP server, tools appear in Copilot agent mode after installation, with no configuration file required; provides a stdio MCP server, allowing P402_API_KEY to be added to the client configuration; provides OpenAI-compatible endpoints that can directly replace the original service by changing only the base URL and API key. The platform operates as "five paths, one protocol," supporting real-time routing of each request to select the optimal model for each call, while users write the same OpenAI-compatible code. P402 also provides features such as an event detail dashboard, evidence package export, financial reports, and CSV appendices, used to track costs, review status, privacy modes, and export verifiable evidence. In addition, the platform includes budget, receipt, and payment tracking for agents, cryptographically constraining each agent's spending and providing receipts and settlement records for each payable AI work item.
In terms of usage, users can start with the free sandbox, run an AI spend audit, view evidence, apply to become a partner, or read the documentation. The platform offers multiple entry points and plan tiers: free sandbox, production plans billed by usage, and enterprise as a sales-led model. Tiers differ in event count limits and data retention periods, for example 25,000 metered events as a hard cap, 250,000 events retained for 30 days, and 2,000,000 events retained for 90 days. Specific pricing and further details are not provided in the material.
P402 also involves payment infrastructure. It is already running in production, settling on the Base mainnet with real USDC, with no testnet or simulation. It supports gasless USDC payments via EIP-3009's transferWithAuthorization, so users do not need to pay gas fees when making payments. The platform has a global Facilitator network with edge deployments in 15 regions and verification latency below 50 milliseconds. It provides real-time cost tracking, transaction history, savings analysis, and multisig treasury security (3-of-5 signature requirement, with an emergency pause function). P402 implements Google Cloud's open Agent-to-Agent (A2A) protocol; the related quotations about this protocol come from public statements by A2A standard partners about the protocol itself, and are not an endorsement of P402.
Applicable scenarios include: teams that want to understand the gross margin of AI features; organizations that need to attribute AI spending by department, employee, workflow, model, and vendor; enterprises that want to set budgets, receipts, and payment tracking for agents; and partners that want to help customers deploy AI spend attributability. P402 also provides Bazaar as a discovery layer for payable endpoints, while Router adds runtime signals so that buyers can choose endpoints based on real health and policy outcomes. The platform emphasizes "give every token an owner," starting from a single metering event and turning token usage into a real-time ledger for budgets, margins, workflows, policy outcomes, and evidence.
In terms of limitations, the materials do not provide clear explanations of features the platform does not support, performance ceilings, compliance certifications, or specific applicable scale. Some features, such as Control runtime flip and Optimize recommendations, are still in a controlled state, with the latter only becoming available after sufficient output data is produced. Regarding pricing, implementation costs, time required for integration, and customer cases, the materials do not currently provide them.
