Founding member pricing — limited spots
Catch your next $34,000 OpenAI bill before it hits.
A cost circuit breaker for OpenAI, Anthropic, AWS Bedrock, and Vercel. One read-only credential — no proxy, no SDK, no prompt access. It polls your spend every 15 minutes and alerts you when a runaway loop starts burning budget.
No proxy.No SDK.No prompt access.One read-only key.
Free spend-risk audit
Which providers do you use?
No proxy·Prompts never leave your stack·Read-only credential by default·Optional revocation uses a separate write-scoped credential·How we protect your keys
Cloudflare Workers
“Durable Object alarm loop. Zero users. No platform warning.”
Read on HN →
Anthropic
“Enterprise-scale bills reveal what indies feel too — no cheap kill switch exists.”
Read on HN →
OpenAI
“Peter Steinberger's OpenAI bill, publicly posted. He's not the only one.”
Read on HN →
Four things ModelBrake does that your provider dashboard won't.
Cloud dashboards show you the damage after the invoice arrives. ModelBrake catches it in minutes and can stop it.
Not a proxy. Nothing routes through us.
You give us a read-only admin key — the one that can see spend but can't call models. Your application keys, prompts, and responses never touch our infrastructure. We can't read them, store them, or slow them down.
Cost polling every 15 minutes
We poll OpenAI, Anthropic, and Vercel AI Gateway usage APIs every 15 minutes; Bedrock cost comes via AWS Cost Explorer every 6 hours. Vercel platform infra (serverless, bandwidth, edge) is monitored via webhook — alerts arrive the moment Vercel sends them. You see the number climbing the same day — not on the invoice three weeks later.
Anomaly detection that isn't a fixed alert
The baseline is the median of your trailing days, so one earlier spike can't poison it. A day running 3× that median fires an alert — even if you're still under your monthly limit.
Spend ceilings and a kill switch
Set a spend ceiling per key. Cross it and you get a critical alert on the next poll. Opt in to revoke-on-breach and it will kill the application key for you — but only with a second, write-scoped credential you grant deliberately, not the read-only key we monitor with.
How ModelBrake compares.
Most cost tools are proxies — they sit between you and your provider, routing every request. ModelBrake never sees a prompt.
| Provider console | Gateway / proxy | ModelBrake | |
|---|---|---|---|
| Setup | Already there | Route all traffic through proxy | One read-only admin key |
| Latency impact | None | Added hop on every call | None |
| Prompt access | Your data stays put | Proxy sees every prompt | Never sees prompts |
| Alert speed | Invoice (days–weeks) | Real-time per request | 15-minute polls |
| Anomaly detection | Manual threshold | Varies by vendor | Automatic — 3× your median day |
| Credential risk | N/A | Production key lives in proxy | Read-only key, separate from app |
Build on your spend data.
ModelBrake exposes a REST API so your internal tools, dashboards, and planning software can pull spend, key, and alert data directly.
GET /api/v1/spend
Read spend snapshots
Pull per-key spend by day across all connected providers. Aggregate it however your tooling needs — weekly rollups, per-project breakdowns, burn-rate forecasts.
PATCH /api/v1/keys
Manage key settings
Update spend ceilings and breach actions from your own interface. Raise a ceiling before a planned spike, lower it after.
GET /api/v1/alerts
Query alerts
Read anomaly and breach alerts into your incident tracker or Slack bot. Filter by severity, correlate with deployment events.
Authenticate with a bearer token you create from your dashboard. The API is read-only by default — write access is limited to key settings you can already change in the UI.
curl -H "Authorization: Bearer mb_live_…" \ https://modelbrake.com/api/v1/spend?days=7
Founding member pricing.
The first 50 accounts lock in at $29/mo forever. When we open general pricing, existing accounts keep their rate.
OpenAI, Anthropic, AWS Bedrock, and Vercel AI Gateway today. Cloudflare Workers next.
Founding member rate, locked for life. Limited spots.
- OpenAI, Anthropic, Bedrock, and Vercel AI Gateway — polled every 15 min (Bedrock every 6h)
- Unlimited keys, unlimited alerts
- Anomaly alerts (3× your median day) + per-key spend ceilings
- Email + webhook delivery (Slack-compatible)
- Every feature we ship next, at this rate — Cloudflare Workers, more providers, and more
- Direct email to us when something's off
$29/mo, billed monthly. Cancel anytime, rate stays locked.