Infrastructure · India · INR settlement

Agents are economic actors.
They need financial rails.

MetricAI sits in front of your model calls. It meters every request, binds spend to agents and sessions, and produces numbers you can bill against. Built for production ledgers, not demos.

Live/Proxy trace

// Route the SDK through MetricAI. One line.
client = metricai.wrap(openai, { project_key: "pk_live__" })

→ POST /v1/chat/completions  intercepted  session=ses_8k2m

₹ figures illustrative — wiring is real.

First 3 Months FREE

Start Free Trial Book Demo

No credit card required

500+

Teams

Production agent workloads tracked

₹10Cr+

AI Spend Tracked

Metered across providers and sessions

99.9%

Uptime

Proxy and settlement infrastructure

T METERING·SESSION ATTRIBUTION·GST-READY LINE ITEMS·PROVIDER COST RECONCILIATION·MCP PAYMENT RAILS·RATE LIMITS BY AGENT·AUDIT-GRADE REQUESTS

The old model breaks under agents.

A static API key and a monthly OpenAI invoice tell you nothing about who burned what. Loops and tool chains turn small mistakes into large bills.

Before

Long-lived API keys. No per-actor limits. Revocation is coarse.
Agents retry and recurse. Spend spikes while product metrics look flat.
Finance sees one provider total. Engineering cannot answer margin per customer.

Agent economy

Every completion flows through a proxy you control. Keys never touch the client.
Usage is tagged with session, agent_id, and end_user.
Rupee-denominated lines roll into invoices your CFO can defend in review.

How traffic flows.

Your application keeps the same SDK surface. MetricAI terminates TLS, normalizes provider quirks, and emits metering events you can bill on.

Your app

Agents, backends, workers. Anything that calls chat or completions today.

MetricAI proxy

Auth, metering, attribution, settlement. One choke point for policy.

LLM providers

OpenAI-compatible and first-party routes. Provider keys stay server-side.

No provider credentials in the browser

Request-level audit trail

Idempotent settlement hooks

Designed for Indian tax reporting workflows

Platform

Everything you need to run AI at scale.

Observability, routing, and billing infrastructure built for India's agent economy.

Cost Tracking

Real-time spend by agent, session, and model. INR-native ledgers finance can reconcile.

Multi-Provider

OpenAI, Anthropic, Bedrock, and more through one proxy. Keys stay server-side.

Smart Routing

Auto-select the best model per request with fallbacks, rules, and A/B testing.

Alerts

Cost, error, and latency thresholds with email, Slack, and dashboard notifications.

Team

Invite members, assign roles, and share dashboards across your organization.

Analytics

Margin analysis, latency trends, tool overhead, and session-level debugging.

Billing

Outcome-based billing.

Tokens are inputs. Customers pay for results. MetricAI keeps both: provider cost from the wire, and commercial outcomes you define. Finance gets a line that maps to a contract clause.

// Bill on completion, not raw tokens

await metricai.record_outcome({

  session_id: "ses_8k2m",

  outcome: "ticket_resolved",

  value_inr: 49.00,

});

Surface vs. depth

One record_outcome call

Token normalization across providers
Streaming completion accounting
Session and agent attribution
Idempotency and replay safety
Ledger events for downstream ERP
GST-ready amount breakdowns

Economics

Agent-level profitability.

Same product surface, different burn profiles. MetricAI rolls provider cost to the agent that triggered it. You see who subsidizes whom.

agent_support_tier1

₹+18,420

Recommendation
Ship more. Margin covers retries.

agent_onboarding

₹0

Recommendation
Breaking even. Cap prompts or shorten the system message.

agent_research

₹−6,900

Recommendation
Disable browse tool or require approval.

Protocol

Stripe for MCP servers.

Tool vendors need pricing that survives partial failures and double posts. MetricAI treats MCP calls like payment intents: correlated, auditable, and priced in INR before they hit your general ledger.

// Agent A pays Agent B for an MCP tool call (sketch)

POST /v1/mcp/transfer

{

  "payer_agent": "agent_sales",

  "payee_server": "mcp_crm_acme",

  "correlation_id": "corr_9x1",

  "amount_inr": "12.50",

  "memo": "lookup_customer_record"

}

Where the agent economy goes.

We are not optimizing for a hero demo. The work is plumbing: identifiers on every call, deterministic metering, and contracts that survive an audit.

India adds GST discipline and rupee settlement expectations. MetricAI is written with that operating context from day one.

Live

OpenAI-compatible proxy

Meter chat and completions. Tag sessions and agents. Export usage for billing.

Building

Settlement pipeline

INR line items, tax metadata, and hooks into invoicing. Fewer spreadsheet bridges.

Roadmap

Cross-agent clearing

Net positions between internal agents and external MCP vendors on one statement.

Roadmap

Policy engine

Hard budgets per agent, circuit breakers on runaway loops, approval gates on spend.

Get started.

Integration is deliberately boring. If you already call OpenAI from the server, you are one wrapper away from metered traffic.

01
Create a project
You get a project key and proxy host. Keys are rotatable without redeploying provider secrets to clients.
02
Wrap the SDK
One wrapper call. Headers carry session and agent identifiers. Streaming responses are still supported.
03
Route traffic
Point staging first. Compare provider invoices to MetricAI totals. Discrepancies should be explainable in one query.
04
Close the loop
Pull usage exports or webhooks into your billing system. Outcome lines are explicit events, not guesses.

pip install metricai-sdk

from metricai import wrap
from openai import OpenAI

base = OpenAI()
client = wrap(base, project_key="pk_live__")

# All completions are metered and attributed
r = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "..."}],
    extra_headers={
        "X-MetricAI-Session": "ses_8k2m",
        "X-MetricAI-Agent": "agent_support_tier1",
    },
)

First 3 Months FREE

Start tracking AI costs in minutes.

Full observability access. No credit card required. Five-minute setup with our SDK.