LLM Cost Estimator
Estimate multi-model API cost from tokens or pasted prompts.
OpenCatch emails, API keys and hardcoded passwords in a prompt before you send it to a model.
Include logs, configs or code you plan to share.
Each match is listed with a type such as email or aws-access-key.
Delete the values or swap them for placeholders.
Confirm the list is empty before sending.
You are about to paste a stack trace and a config snippet into a chatbot to debug a failing deploy. Somewhere in that wall of text there might be a live key. Prompt PII Scanner checks for the obvious ones first.
Paste the prompt and the findings list updates as you type. Each finding is labeled with its type and the matching text. The scanner looks for email addresses, keys that begin with sk, rk or pk plus a hyphen or underscore and at least 16 letters or digits, Google API keys starting with AIza, GitHub tokens starting with ghp_, AWS access key IDs starting with AKIA, and assignments such as password=, passwd: or secret=. It reports up to 50 findings, or a clear message when nothing matches.
It is a fast first pass with known gaps. It does not look for phone numbers, card numbers or names, and keys with extra hyphenated segments after the prefix may not match. For broader personal data, run the text through the PII Detector. The scanner only reports; to replace findings with placeholders, use Secrets Redactor for credentials or the On-Device NER Redactor, which also masks names and places.
Scanning runs on the text inside this tab. The point is to find secrets before they leave your machine, so the tool itself never sends them anywhere.
Keyboard shortcut
Answers for this tool. For site-wide help, open the FAQ hub.
Remove it from the prompt, and if it was ever pasted into a third-party tool or committed anywhere, rotate it with the provider. Deleting it from your text does not revoke a key that has already leaked.
Patterns are narrow by design to avoid noise. Keys from other providers, keys with additional hyphenated sections after the prefix, and tokens without a known prefix will not match. A clean result is not proof that the text is safe.
Yes, any string shaped like an email is listed, including example.com addresses and noreply senders. Decide case by case whether an address is personal data in your context.