CostVane

Privacy-first AI cost control

Know what every AI call will cost before it runs.

Estimate requests, enforce budgets, and attribute actual spend to the customers and features creating it—without routing model traffic through another gateway.

Make your first estimate Become a design partner
  • No prompts
  • No outputs
  • No provider keys
POST /v1/preflight
DECISIONallow
Estimated cost
$0.0064
Budget remaining
$182.41
Pricing source
official · fresh

One control layer across

OpenAIAnthropicGoogle GeminiOpenRouter

The margin problem

Your AI bill says what you spent. It does not tell you what to stop.

Provider dashboards arrive after the cost. Proxies add architectural and privacy overhead. CostVane sits beside your stack and gives each request a price, owner, and policy before the bill becomes a surprise.

01

Unknown unit economics

See cost by customer, feature, workflow, team, project, or environment.

02

Reactive budget alerts

Return allow, warn, or block before an expensive model request executes.

03

Fragmented providers

Use one normalized API across four major AI model ecosystems.

Three calls. One feedback loop.

Estimate. Decide. Reconcile.

Start with a single endpoint, then add policy and attribution as your product grows.

1
POST /v1/estimate

Price the request

Calculate input, cached-input, output, and supported long-context costs from a versioned catalog.

$0.0064 estimated
2
POST /v1/preflight

Apply the guardrail

Evaluate dollar or token budgets scoped to the customer, feature, workflow, team, or project.

allow · 72% remaining
3
POST /v1/usage

Record reality

Ingest provider-reported cost idempotently and reconcile actual versus estimated spend.

variance +$0.0003

Privacy by omission

Cost control without becoming your model gateway.

CostVane needs token counts and business dimensions—not the sensitive content flowing through your AI product.

Read the privacy policy
DataCostVane
Token countsUsed
Customer / feature IDsOptional
Prompts and outputsNever needed
Provider API credentialsNever needed
Proxied model trafficNever needed

Built for production economics

The controls between usage and margin.

↗

Fresh pricing provenance

Official source URLs and freshness dates travel with every estimate.

◎

Actual-cost reconciliation

Compare your estimate with provider-reported cost and measure variance.

⊞

Multidimensional attribution

Group spend by customer, feature, workflow, team, project, environment, or tags.

◇

Scoped budgets

Enforce dollar or token caps for the exact part of your product that needs a guardrail.

≡

Idempotent ingestion

Safely retry usage events without double-counting spend.

⌁

Signed webhooks

Receive HMAC-signed budget events with encrypted webhook secrets at rest.

Simple launch pricing

Cost control should cost less than the waste it prevents.

Every plan includes all four providers, privacy-first architecture, estimation, attribution, and budget enforcement.

Free

$0/ month

Explore the API and validate your integration.

  • 5,000 operations
  • All supported providers
  • Community support
Start building
Best for builders

Builder

$19/ month

Ship cost visibility into an early-stage AI product.

  • 100,000 operations
  • Budgets and webhooks
  • Email support
Request access

Growth

$49/ month

Control unit economics across growing workloads.

  • 500,000 operations
  • All attribution dimensions
  • Priority support
Request access

Custom

Let’s talk

Tailored limits and onboarding for larger teams.

  • Custom operations
  • Integration support
  • Commercial terms
Contact us

Proposed overage: $5 per additional 100,000 operations. Launch pricing may evolve with design-partner feedback.

Frequently asked

Clear answers before your first call.

Is CostVane another LLM proxy?

No. Your application continues calling model providers directly. CostVane only receives the cost and attribution metadata you choose to send.

Do you store prompts or model outputs?

No. CostVane’s API is designed around token counts, provider/model identifiers, actual cost, and optional business dimensions.

Where does pricing come from?

The versioned pricing catalog records official source URLs and freshness dates for OpenAI, Anthropic, Google Gemini, and OpenRouter.

Can I block a request before it runs?

Yes. The preflight endpoint evaluates your scoped budgets and returns an allow, warn, or block decision before you call the model.

Design partners

Turn AI spend into a product decision.

Get 30 days of Growth access, direct onboarding, and a voice in the first SDK.

Apply for the program Best for AI SaaS, agencies, and agent teams.