Free betaFor builders running AI agents on OpenAI, Anthropic and OpenRouter
Find out about a runaway AI bill from us, not from your invoice.
KeyBark watches your API spend from our servers and flags it when an agent loop or a leaked key starts burning money. Alerts go to your email and your dashboard. Optional kill switch (beta). No proxy, no SDK, no browser tab to keep open.
Free plan covers one key. No card needed.
This keeps happening
First-hand posts from builders, and news reports, from the last few weeks. Click through to read them.
in unauthorized charges overnight, reported by some users after the OpenCode Console key breach (BleepingComputer)
Source (opens in a new tab)of image generation in 5 hours on an OpenAI key stolen in Railway's Oct 21 env-var breach
Source (opens in a new tab)API bill from an infinite loop in a CSS file
Source (opens in a new tab)of charges an hour after setting a $30 OpenAI spend limit (Hacker News)
Source (opens in a new tab)in unauthorized usage on a hacked API key, despite a $40 Anthropic spend limit
Source (opens in a new tab)- OpenAI spend alerts only notify; a cap needs “Enforce a hard limit”.
- Anthropic can’t cap the Default Workspace.
A free yes/no check of your OpenAI, Anthropic and OpenRouter settings. No sign-up, no key needed. Or watch one key free — alerts by email and on the dashboard, about every 15 minutes during the beta.
How it works
→ 1. Connect a key
Paste an OpenAI Admin key, an Anthropic Admin key or an OpenRouter key. We validate it against the provider and encrypt it before storing it.
→ 2. Set your thresholds
Pick a daily and a monthly dollar threshold and turn on spike alerts (today vs. your 7-day average).
→ 3. Get warned (or stopped)
When a threshold is crossed, an alert is emailed to you and appears on your dashboard. If you opted in to the kill switch (beta), KeyBark also disables the key you named through the provider's API.
Checks run about every 15 minutes during the beta, plus on demand with “Check now”. Supported today: OpenAI (organization Costs API), Anthropic (Admin cost report) and OpenRouter (key usage). Integrations are marked beta until they have been verified against many real accounts.
Leak watch: about once a day we also search public GitHub code for the key you connected, and send an urgent alert with a link to the page if it shows up. It can only find keys exposed in public places that GitHub’s code search has indexed (not private repos, git history, chats or other sites), and KeyBark does not revoke or rotate keys for you.
Built like we hold your keys, because we do
Alert-only by default
New keys are only read. Nothing is ever disabled unless you turn on the kill switch for that key.
Encrypted at rest
Provider keys are encrypted with AES-256-GCM before storage. The encryption key lives in a server environment variable, never in the browser or the code repository.
Least privilege
Use the narrowest key the provider offers. You can delete a key from KeyBark at any time, and the encrypted copy is deleted with it.
No proxy
Your production traffic never passes through us. We only call the providers' billing and key-management endpoints, plus GitHub code search for leak watch (with a short fragment of the key, never the whole key).
Pricing
Start free with one key. Upgrade when you have more keys to watch or want the kill switch (beta).
One key, watched for you.
- 1 provider key (OpenAI, Anthropic or OpenRouter)
- Daily, monthly and spike alerts by email and on your dashboard
- Leak watch: daily search of public GitHub code for the key
- 30-day spend chart
- Free limit audit + leak response plan
For solo builders running agents.
- Up to 3 provider keys
- Everything in Free
- Opt-in kill switch (beta; not yet tested on live provider accounts)
For small teams with several projects.
- Up to 15 provider keys
- Everything in Indie
- Project or client label on each key
For agencies managing client keys.
- Up to 50 provider keys
- Everything in Pro
- Client labels on every key and alert
KeyBark is in beta. Prices in USD per month, billed monthly in advance through Stripe until you cancel. Cancel any time from the Stripe billing portal; your plan stays active until the end of the period you paid for. By subscribing you agree to our Terms and Privacy Policy. Built today: email and dashboard alerts, leak watch (a daily search of public GitHub code for your key, with a rotation checklist) and the opt-in kill switch (beta). Not built: Slack, Telegram and SMS alerts; faster polling; leak scanning beyond public GitHub code (private repos, git history, gists, paste sites); automatic key revocation or rotation; any emergency or incident-response service.
FAQ
Doesn't OpenAI already have spend limits?
It has two different things. A spend alert only sends a notification and traffic continues. A hard limit stops traffic only if you turn on “Enforce a hard limit”, and OpenAI says enforcement is not instantaneous. KeyBark adds an independent watcher on top, plus alerts for spikes, and helps you check your settings with the free limit audit.
What about Anthropic?
Anthropic lets you set spend limits per workspace, but not on the Default Workspace. Its usage and cost data usually appears within about 5 minutes. KeyBark reads the Admin cost report and alerts on your thresholds, including Default Workspace spend.
What key do I give you?
For OpenAI, an Admin key (it can read organization costs). For Anthropic, an Admin API key (organization accounts only). For OpenRouter, the API key you want to watch, or a provisioning key plus the target key's hash if you want the kill switch. Use the most limited key the provider allows.
How often do you check?
Right now checks run about every 15 minutes during the beta, plus on demand with “Check now”. Provider cost data also lags (Anthropic says about 5 minutes; OpenAI's cost data is reported in daily buckets), so KeyBark is an early-warning system, not a real-time firewall.
Will the kill switch take my app down?
Only if you turn it on. It is off by default, needs an explicit confirmation, and only fires when a daily or monthly threshold you set is crossed. It is in beta: test it on a non-production key first. Spike alerts never trigger it.
How is KeyBark different from similar tools?
Start with your provider's own controls: they are free and they are the only limits enforced at the provider. OpenAI has project and organization hard spend limits, Anthropic has workspace spend limits (not on the Default Workspace), and OpenRouter has per-key credit limits plus a low-balance email. Independent watchers include BillCurfew (OpenAI, Anthropic and OpenRouter, checks about every 5 minutes, Slack and auto-kill on paid plans), ModelBrake (OpenAI, Anthropic, AWS Bedrock and Vercel, about 15-minute checks) and API Monitor at apimonitor.ai (many providers, keys stay in your browser, polls while its tab is open). Gateways such as Portkey enforce budgets because your traffic passes through them. KeyBark checks about every 15 minutes from its own servers with no proxy, adds a daily leak watch on public GitHub code, has a free one-key plan and an opt-in kill switch (beta). If you need Slack alerts or faster checks today, one of the others may fit better.
Do I need to install anything?
No. There is no proxy, SDK or browser extension. KeyBark talks to the providers' billing APIs from our servers, so nothing has to stay open on your laptop.
Do you detect leaked keys?
Partly. Leak watch searches public GitHub code for the key you connected about once a day and alerts you urgently if it appears, with a link to the page and a rotation checklist. It cannot see private repositories, git history, gists, chat messages, pastes or other sites, so it only catches keys exposed in public places GitHub has indexed. It's an early warning, not an incident response service: revoking and replacing the key is up to you.