Which metrics should be recorded?
Record request count, success rate, status category, timeout rate, and latency percentiles by endpoint and service. Keep IP addresses and response data out of general metrics labels to avoid sensitive high-cardinality telemetry. Redact all authentication headers.
How should capacity alerts work?
Create thresholds for account remaining_requests and important key remaining_requests. A Pro key can exhaust its own cap while the account still has room. Include the key's descriptive operational identity in alerts without exposing the secret or publicizing internal usage data.
What makes an alert actionable?
State the affected service, endpoint, status, observed usage, expected owner, and immediate fallback. Link to a runbook for rotation, plan review, and retry suppression. Review alert thresholds after launches and seasonal events so expected growth does not hide genuine loops or credential abuse.
How should this be operated in production?
Track request count, status category, timeout rate, and latency percentiles by endpoint and named service. Poll GET /usage at a modest interval to compare shared account capacity with individual key consumption. Alert before remaining requests reach zero, document ownership, and maintain a fallback for optional enrichment. Avoid metrics labels containing raw addresses, credentials, or other sensitive high-cardinality values.
How should capacity and retries interact?
A Pro key can reach its configured cap while shared account capacity remains, and the account can reach its plan allowance before another key reaches its maximum. Diagnose both levels before changing traffic. HTTP 429 should suppress aggressive retries because repeated calls cannot restore capacity. Bound worker concurrency and account for traffic from staging, tests, background jobs, and browser integrations in the same monthly total.
How should problems be diagnosed step by step?
Reduce the integration to one authenticated cURL request from trusted infrastructure, using GET /ip/8.8.8.8 as the known explicit fixture. Record the HTTP status, content type, elapsed time, and safely redacted body. If that succeeds, compare the application URL, headers, proxy behavior, timeout, and JSON model. If it fails, separate authentication, validation, capacity, transport, and server categories before changing code. Check GET /usage after 429, compare exact browser Origin values after CORS failure, and confirm the deployed environment received the intended secret. Do not rotate, retry, or increase limits blindly; preserve enough controlled evidence to identify the boundary that actually failed.
How should this behavior be tested?
Create unit fixtures for successful lookup or usage data, nullable optional fields, unknown additive properties, and every relevant error status. Mock timeouts and malformed intermediary responses so the client never mistakes an HTML proxy page for valid JSON. Run a small integration test with a designated nonproduction key and 8.8.8.8, inject the secret through CI, and keep live calls out of ordinary unit-test loops.
What privacy and accuracy boundaries apply?
IP intelligence is approximate network context. It does not prove identity, residence, billing country, precise device position, or individual ownership of a network organization. Retain only the source and derived fields required for the product purpose, restrict access, define deletion, and let users correct meaningful defaults. Security, financial, legal, or physical-location decisions require additional verified evidence and appropriate professional review.
Which plan and usage limits apply?
The 7-Day Developer Key provides 1,000 evaluation requests. Free includes 10,000 monthly requests and 1 active key. Starter includes 100,000 requests, 2 active keys, and up to 5 exact CORS origins per key. Pro includes 1,000,000 requests, 5 active keys, up to 10 origins per key, and a configurable 100 to 1,000,000 cap at key creation. Active keys share the account allowance.
| Plan | Monthly requests | Active keys |
|---|---|---|
| Free | 10,000 | 1 |
| Starter | 100,000 | 2 |
| Pro | 1,000,000 | 5 |
What should happen before release?
Confirm the endpoint and response model, validate URL construction, keep the secret outside source, set timeout and status-aware retry behavior, and test the real deployment environment. Add monitoring for latency, errors, shared usage, and per-key consumption. Document the owner, purpose, expected volume, fallback, retention, and rotation procedure. Review the implementation again whenever traffic, hosting origins, plan limits, or product decisions materially change.
External references
Continue with the standards and official documentation most relevant to this guide.