policy applied · model rerouted identity: cust_acme

RUNTIME USAGE CONTROL

Measure every interaction.
Control every resource.

Identify the customer behind every request, measure what they consume, and enforce your business policies while the request is still running.

Read the docs

@usageflow/vibe Python

usageflow-vibe

Go · Vibe

OpenAI •

Anthropic

Metering tells you what happened. UsageFlow decides what happens next.

Free: 1 workflow with a live policy • No credit card

Scroll

Runtime control

Decide what runs, while it runs.

Your code says what it wants. UsageFlow checks who is asking and shapes what actually executes.

Illustrative example · customer-support · identity cust_acme · Console-edited policy, no redeploy

Requested

Modelopus-4
ReasoningHigh
ToolsOn
Tokens8k

UsageFlow Awaiting

Executed

                01 intent
                
                  Recording intent
                  usage.chat() · opus · reasoning=high · tools · 8k
                
                02 policy
                
                  Checking policy
                  budget-guard · identity cust_acme at $98 / $100 cap
                
                03 transform
                
                  Rewriting plan
                  sonnet · reasoning=low · tools off · 4k tokens
                
                04 why
                
                  Explaining change
                  budget exceeded · budget-guard · no redeploy

Developer sends intent — Console policy has not run yet.

Policy-driven execution

Write intent once. Console reshapes the next call.

Console policies rewrite the execution plan — model, tools, budgets — before the provider runs. Edit the policy in the Console, no redeploy. Illustrative sample data.

This request · no redeploy

Requested

opus · high · tools · 8k

Executed

—

Reason — Policy —

✓ Plan rewritten at runtime
✓ Traces show why — no redeploy

See Documentation →

Why UsageFlow

Your customers’ usage is connected. Your infrastructure isn’t.

Today

Authknows the customer

Logsknow the usage

Codeknows the rules

Nothing connects them.

With UsageFlow

01
Identifyidentity: 'cust_acme'
02
Measure1,284 requests · $42.18
03
Controlallow · transform · block

one identity · one ledger · one policy layer

Metering tells you what happened.
UsageFlow decides what happens next.

Policy engine

Policies decide what happens next.

Rules on who the customer is, their plan, and what they’ve used. Checked on every request.

Trigger

WhenChat request comes inworkflow = customer-support

Tiers first match wins

usage reaches$100Cost ($)

Then

Block the request

usage reaches$95Cost ($)matched · cust_acme

Then

Route to a different modelclaude-sonnet-5

Effects

Notify SlackReport to a Stripe meter

Example policy

ConditionsIdentity, plan, and metadata you send.
ThresholdsTokens or spend, per customer.
TiersDifferent outcomes at different levels.
ActionsRoute, degrade, or block (Vibe SDK).
EffectsSlack, webhook, Stripe meter.

Not just models.

The same customer identity and ledger cover more than LLM calls.

AI modelsRequested vs executed model, per customer (Vibe SDK).
APIsUsage per customer on your endpoints, through API middleware.
WorkflowsUsage grouped by the workflow you name on each call.

Identity · usage · cost

Every interaction belongs to a customer.

Usage, cost, and policy decisions, tied to the customer behind them.

EXAMPLE LEDGER sample data

CUSTOMER

cust_acme

Acme Inc.

USAGE

1,284

requests

SPEND

$42.18

this month

BILLING

Stripe

Synced

16.6% below requested cost this month · sample data

Integration

Integrate through code. Control through policy.

Pass identity and workflow with each call. Policy lives in the Console.

TypeScript

import { usage } from '@usageflow/vibe';

usage.init({ apiKey: process.env.USAGEFLOW_API_KEY! });

const result = await usage.chat({
  // Identity → scoped ledger (your customer, not an API key)
  identity: 'cust_acme',
  workflow: 'customer-support',
  provider: 'anthropic',
  model: 'claude-sonnet-5',
  messages: [{ role: 'user', content: prompt }],
});

✓ identity binds to the ledger ✓ Console may rematch model / tools / budget ✓ Trace shows requested vs executed

View TypeScript setup guide →

Python

from usageflow.vibe import VibeClient, Message

client = VibeClient()  # reads USAGEFLOW_API_KEY

result = client.chat(
    # Identity → scoped ledger (your customer, not an API key)
    identity="cust_acme",
    workflow="customer-support",
    model="claude-sonnet-5",
    messages=[Message("user", prompt)],
)

✓ Same identity and workflow fields ✓ Synchronous client; provider inferred from model

View Python setup guide →

import "github.com/usageflow/usageflow-go-middleware/v2/pkg/vibe"

client, err := vibe.New(vibe.Options{APIKey: os.Getenv("USAGEFLOW_API_KEY")})
if err != nil {
	log.Fatal(err)
}
defer client.Destroy()

res, err := client.Chat(ctx, vibe.ChatRequest{
	// Identity → scoped ledger (your customer, not an API key)
	Identity: "cust_acme",
	Workflow: "customer-support",
	Provider: vibe.ProviderAnthropic,
	Model:    "claude-sonnet-5",
	Messages: []vibe.Message{{Role: "user", Content: prompt}},
})

✓ Same identity and workflow fields ✓ Explicit client lifecycle: New / Destroy

View Go setup guide →

FAQ

Straight answers for developers.

What is UsageFlow?

UsageFlow is a runtime usage control layer. Your app passes a customer identity and workflow with each call; UsageFlow measures consumption, attributes it to that customer, and applies Console policies while the request runs. With the Vibe SDK for AI calls, policies can enforce budgets and route models. The Vibe SDK is the current integration for AI calls: usage.chat() and usage.prompt().

How is this different from an AI gateway?

Gateways sit on the provider HTTP path and mostly route and log by API key. UsageFlow runs inside your app, merges Console policy with your intent, transforms execution when needed, meters what actually ran, and can sync executed usage to Stripe, on identity and workflow dimensions. Metering tools tell you what happened; UsageFlow also decides what happens next.

How does runtime policy evaluation work?

On each call, UsageFlow identifies the customer, reads their current usage, and evaluates the matching Console policy (thresholds, tiers, plan or metadata scope) before the provider runs. The outcome is to allow, route to another model, or block, and the trace records the reason.

Do Console policy changes require a redeploy?

No. Policy edits in UsageFlow Console apply on the next call. Traces show requested vs executed settings and the reason for any transform.

How does customer identity and attribution work?

Pass your customer or tenant id as identity. UsageFlow scopes a ledger to that identity and can report (Console / Stripe) on identity and workflow dimensions, so spend reads in language your customers understand, not just API keys.

Which SDKs and providers are supported?

The Vibe SDK for TypeScript, Python and Go. It targets OpenAI and Anthropic through a unified API. For full provider-specific surfaces, use usage.wrap(nativeClient) alongside the unified path.

What's included on the free plan?

Free includes one workflow with live policy (basic model ladder; policies edited in the Console apply without redeploy). Paid tiers add more workflows, tool gates, generation caps, and Stripe sync.