League SoftwaresAI SHIELD

IN DEVELOPMENT · PILOT TEAMS WELCOME

Know what is
using your
AI keys.

League AI Shield watches your OpenAI and Anthropic API usage and spend, alerts you when it changes unexpectedly, and helps you shut off a key that may have leaked. We are building it now with a small group of pilot teams.

Join the pilot What it can't see

OpenAI and Anthropic first · No prices yet · Nothing to install today

What each mode can see

MonitorProtect
Spend per dayProvider reportedEstimated live
Spend per hourNot reportedEstimated live
Tokens per key and modelYesYes
Request countsOpenAI onlyYes
Caller's networkNeverApproximate
Stop a keyWhere the provider allowsInstantly, at the gateway

Monitor reads your provider's admin reports. Protect means your app sends its AI calls through our gateway.

THREE STEPS

See it. Get told.
Contain it.

What we are building, for developers, AI startups and agencies. Monitoring and alerts come first; the gateway follows.

01 / MONITOR

Every key, model and dollar in one place.

Connect an OpenAI or Anthropic admin key. We read usage and cost reports, list your API keys and show how fresh the data is.

  • No changes to your app
  • Costs labelled as reported or estimated
  • Read-only by default

02 / ALERT

Your own alert rules, checked against history.

Set thresholds or "three times the usual for this hour of the week", then replay the rule over past data to see how often it would have fired.

  • Minimum amounts to avoid noise
  • Refuses rules your data can't answer
  • Email and webhook first

03 / CONTAIN

Pause a key without guessing.

Block a key at our gateway in one click. Disabling or deleting it at the provider needs a person to approve it, and we only call it done once the provider confirms.

  • Every action in an audit log
  • AI assistants can ask, not act
  • Clear "do this by hand" steps when needed

STRAIGHT ANSWERS

What it can't see.

A security tool should say where its view ends.

Who used a stolen key

Provider reports name the key and model, never the person, device or IP address. Only traffic sent through our gateway shows a network, and even then location is approximate.

Traffic that skips the gateway

Blocking a key at the gateway stops calls that go through it. A copy of the original provider key used directly keeps working until it's disabled at the provider.

Per-minute spend from reports

OpenAI and Anthropic report costs once per day. Hourly spend alerts need gateway traffic, so we won't offer them on reports alone.

Proof of theft

An unusual spike is a finding to investigate, not proof. A deploy bug or a retry loop can look the same, so we never revoke a key on a score alone.