AI Cost Circuit Breaker Pro 1.5

Stop the runaway pattern
before it consumes the month.

Retry loops, repeated requests, oversized input or output, an expensive model, or a background job can turn one bug into a fast-growing bill. Pro limits the blast radius by time, request, caller, model, and execution context.

Monitoring firstRaw prompts never persistedOne production hostname
Measure a baseline with Free →
GUARDRAILS · ENFORCINGPRO 1.5
CRITICAL

Retry circuit opened

request_retry_circuit
Source
inventory-sync
Resolved model
openai / gpt-5
Context
background
Matching failures
5 / 5
DecisionBlocked before provider call
Illustrative Pro 1.5 protection record
Why Pro exists

Runaway spend begins in seconds,
not at month end.

A monthly ceiling is a necessary last wall. Pro adds earlier, cause-specific decisions so one malfunction does not get the entire budget to itself.

01 · BACKGROUND JOB

A Cron task retries every failed 5xx response.

The same failure keeps opening a new provider call while nobody is watching the dashboard.

Background scope + matching-retry cooldown
02 · PUBLIC CHAT

The same request is submitted again and again.

A form loop or automated visitor repeats the same normalized work faster than a monthly counter explains it.

One-minute velocity + keyed repetition detection
03 · MODEL CHANGE

A plugin update resolves to a costlier model.

Total usage can look ordinary while estimated spend changes sharply for one caller and one model.

Per-source budget + provider/model cap
Protection architecture

Three layers.
One deliberate order.

Deterministic safety rules run before business-aware signals. An absolute hard stop remains the final authority.

LAYER 01

Cause-based guardrails

Control velocity, request size, repetition, retries, caller, model, and execution context with explicit thresholds.

LAYER 02

Smart judgment

Use Human, Bot, Unknown, and optional WooCommerce or custom revenue signals to avoid unnecessary interruption of valuable demand.

LAYER 03

Operations & evidence

Receive threshold and incident alerts, keep blocked reasons, export records, and see a recommended response without storing raw conversations.

Seven Pro 1.5 guardrails

Control the cause,
not only the monthly total.

Every new guardrail is opt-in. Start in Monitoring, review normal behavior, disable or adjust any unsafe rule, and only then switch the global mode to Enforcement.

01
Token velocity

Intervene during the surge

Rolling one-minute and one-hour limits include conservative reservations for in-flight calls.

02
Per-source budgets

Contain one caller's accident

Set monthly USD, request, or token budgets for a plugin, mu-plugin, theme, core, or unknown source.

03
Repeated requests

Recognize the loop

Match canonicalized work using a keyed HMAC-SHA256 digest without persisting raw prompt content.

04
Per-request input & output

Bound one oversized call

Block an estimated input that is too large and explicitly block or clamp supported output limits.

05
Retry storms

Stop the retry chain

Cool down matching retryable failures and use a source-level fallback circuit when exact transport matching is unavailable.

06
Provider/model spend

Separate expensive models

Apply a monthly estimated-USD cap to the resolved provider and model using rates you configure.

07
Execution context

Choose where blocking applies

Apply guardrails to public, signed-in, administration, or background/Cron/CLI requests.

Free versus Pro

Free protects the monthly ceiling.
Pro limits the blast radius.

Free is a dependable sitewide baseline. Pro is for production sites where the cause, speed, and business value of a spike matter.

ControlFreePro 1.5
TimeSitewide monthly request and token boundariesOne minute, one hour, and one request
ScopeSitewideCaller, provider/model, and execution context
CauseAggregate usageRepetition, retries, oversized input/output
USDEstimated monitoringSite, source, and provider/model enforcement
DecisionFixed hard stopDeterministic guardrails + Smart signals + absolute stop
OperationsLocal events and CSVThreshold alerts, incidents, recommendations, Slack/webhook, blocked audit
After the circuit opens

Stop it. Explain it.
Prepare for the next one.

A block without context creates another emergency. Pro records the operational reason, inferred source, resolved model, context, and conservative usage metadata locally.

50 / 75 / 90 / 100%

Email thresholds for sitewide and per-source limits

INCIDENT HISTORY

Traffic confidence, velocity, action, and recommended response

EMAIL · SLACK · WEBHOOK

Optional notifications with an allowlisted, content-minimized payload

BLOCKED AUDIT · CSV

Review and export operational evidence without request fingerprints

Recent protection decisionsLOCAL
UTCSOURCECONTEXTREASON
02:13:18inventory-syncbackgroundrequest_retry_circuit
02:13:11support-chatpublicduplicate_request_loop
02:12:46product-writeradminmodel_monthly_usd
✓ Raw prompt and response content is not displayed or stored
Safe rollout

Observe first.
Enforce when ready.

Pro is designed to avoid turning a new safety setting into a new outage.

01

Run supported native AI Client calls and identify normal sources and models.

02

Add current input and output rates when you need estimated-USD policies.

03

Enable only the relevant guardrails and leave the global mode in Monitoring.

04

Review would-block decisions, disable or adjust unsafe rules, then switch the global mode to Enforcement.

05

Test email delivery; save optional Slack/webhook settings only after checking the destination and payload. Enable Smart Protection when business signals help.

Open the setup guide →

The guard watches operational patterns—not stored conversations.

Raw prompts and responses are never persisted. When repetition protection is enabled, selected request fields are canonicalized transiently in memory and only a keyed, non-reversible HMAC-SHA256 digest and operational counters are stored locally.

Privacy details →

Know exactly what is protected.

Token and USD values can be conservative or incomplete estimates. The provider invoice remains authoritative.

FULL

Default guarded AI Client transporter

Lifecycle accounting plus exact matching-request/retry checks, resolved-model caps, estimated-input checks, and provider-aware supported-output block or clamp before send.

CORE-ONLY

Custom or sidecar transporter

Lifecycle accounting and source-level fallback protection; exact second-stage transport controls may be bypassed.

OUTSIDE

Direct provider SDK or HTTP call

Calls that bypass the native WordPress AI Client lifecycle are outside the current product scope.

One production hostname

Seven guardrails plus
Smart Protection.

Annual billing is the best value. The same complete Pro feature set is also available month to month.

SAVE $21 / YEAR
PRO · ANNUAL
$99 / year
$8.25/month equivalent
  • ✓ All seven Pro 1.5 guardrail groups
  • ✓ Smart Protection and optional revenue signals
  • ✓ Threshold and incident notifications
  • ✓ Blocked-call audit, retention controls, and CSV
  • ✓ Authenticated Pro updates and support
Subscribe — $99/year →

$99 is charged now and every year until canceled online. One production hostname.

Estimated USD uses rates you configure and is not a provider billing cap. Keep provider-side limits as the final backstop.

Questions before activation

Clear scope.
No safety theater.

Does Pro replace provider-side limits?+

No. Provider limits remain an important final backstop. Pro adds earlier WordPress-side decisions by caller, context, request pattern, and resolved model.

Can Pro guarantee that no customer is blocked?+

No. Classification is a signal, not certainty, and deterministic or absolute limits can still stop a request. Smart Protection aims to avoid unnecessary interruption of valuable demand.

Does repetition protection store prompts?+

No raw prompt or response is persisted. Selected fields are processed transiently and only a keyed HMAC-SHA256 digest is stored locally when the feature is enabled.

What happens if the subscription expires?+

Saved monthly request, token, and USD limits plus the seven guardrails continue. Smart traffic and revenue decisions, settings changes, early 50/75/90 alerts, seller-hosted downloads and updates, support, and license services pause until access is restored. The 100% alert, emergency stop, CSV export, and data deletion remain available.

Is the estimated cost my invoice?+

No. It uses configured model rates and recorded or reserved usage. The provider invoice is authoritative.

Pro 1.5 · one production hostname

Do not give one broken loop
the whole month's budget.

Protect one production hostname — $99/year →