Replace IP addresses, emails, hostnames, and API keys in your logs with consistent tokens before you paste them into an AI chat. Paste the reply back to restore the real values.
This tool runs entirely in your browser. What you enter is never sent anywhere.Your input is never sent anywhere.
DetailsPaste your log
What to anonymize
Replacement style
Paste a log and click “Anonymize log”. The result and the mapping table will appear here.
Detection is automatic and may miss some values. Always read the output before you share it.
How to Anonymize Log Files Before Pasting Them Into AI
An error log is the quickest thing to paste into an AI chat, and one of the riskiest. A single stack trace can include client IP addresses, customer emails, session IDs, internal hostnames, and sometimes API keys or passwords.
This tool finds those values and replaces each one with a token. The same value always gets the same token, so an IP address that appears three times becomes IP_1 three times. The AI can still see which events are related, but it never sees the real data.
The tool keeps a mapping table in memory. Paste the AI’s answer into the restore box and the tokens turn back into your real values. Everything runs in your browser, and the log you paste is not sent to our servers.
When Redacting a Log Makes Sense
Debugging a production error with an AI assistant
Strip customer IPs, emails, and session IDs from a stack trace before you paste it into ChatGPT or a similar tool.
Posting a log in a GitHub issue or forum thread
Public posts get indexed and copied. Replace hostnames and credentials first so you can share the log without exposing your setup.
Sending logs to a vendor or contractor
Show an outside team the shape of the problem without handing over real user data or internal server names.
Working under a no-raw-data policy
If your team does not allow customer data in outside tools, redact first and keep the mapping table on your own machine.
How to Redact a Log File in 4 Steps
- Paste your log into the input box, or click Load sample to try it.
- Choose what to anonymize: IP addresses, emails, UUIDs, hostnames, secrets, or phone numbers. Add up to five custom regex patterns for internal IDs.
- Pick a replacement style, then click Anonymize log. You get the redacted log and a mapping table.
- Copy the redacted log into your AI chat. When the answer comes back, paste it into the restore box to put the real values back.
Note: Your log and the mapping table stay in your browser’s memory. They are cleared when you reload or close the page.
Limits of Pattern-Based Detection
- Detection is pattern-based. It can miss unusual formats and flag harmless text, so read the output before you share it.
- Names, street addresses, and other free text are not detected. Add a custom pattern for any identifier you need to hide.
- Hostnames are detected only when they end in a common domain suffix (for example .com, .net, .org, .io, .dev) or an internal suffix such as .local, .internal, .lan, or .corp. A bare name like db01 is not detected.
- Version strings that look like IPv4 addresses, such as 1.2.3.4 or Chrome/120.0.0.0, can be flagged as IP addresses.
- The mapping table and its CSV contain your real values. Treat them like the original log.
- Running the tool again with different input or settings replaces the mapping, so token numbers can change.
What Gets Detected and What It Becomes
Built-in detectors and the token each one produces in token mode.
| Data type | Example | Becomes |
|---|---|---|
| IPv4 address | 203.0.113.45 | IP_1 |
| IPv6 address | 2001:db8::1 | IP_2 |
| Email address | [email protected] | EMAIL_1 |
| UUID / session ID | 550e8400-e29b-41d4-a716-446655440000 | UUID_1 |
| Hostname | web-prod-01.example.com | HOST_1 |
| Secret (value only) | password=hunter2xyz | password=SECRET_1 |
| Phone number (US, off by default) | (555) 123-4567 | PHONE_1 |
IPv4 and IPv6 addresses share one counter, so numbering continues across both. For secrets, only the value is replaced and the label stays. Phone detection is off by default because digit patterns often match order numbers and other harmless text.
Blacking Out vs. Consistent Tokens: How Log Pseudonymization Works
Why raw logs are risky to share, and why this tool replaces values with consistent tokens instead of deleting them.
Why raw logs are risky to paste into an AI chat
Logs often contain IP addresses, session IDs, user emails, and internal server names, and sometimes credentials that ended up in a URL or a debug line.
Sending that to an outside service may conflict with your company’s security policy, a customer contract, or privacy obligations. Removing the sensitive values first takes a minute and avoids the question.
Consistent tokens keep the log readable
If you replace every value with the same block of characters, you lose the pattern, such as one IP address hitting the same endpoint over and over.
This tool gives each new value a numbered token (IP_1, IP_2, and so on) and reuses that token whenever the value appears again. The AI sees the same structure as your real log, with none of the real identifiers.
Reversible redaction: restore the AI’s answer
Each run builds a mapping table of tokens and original values.
When the AI replies with something like “IP_1 is the likely cause,” paste the reply into the restore box and the tool swaps the tokens back to your real values. Restoring works in token mode only, because partial masks can be shared by different values.
Why automatic redaction still needs a manual check
Detection uses regular expressions, so values in unusual formats can slip through, and harmless text can be replaced by mistake.
Treat the tool as a first pass that saves you time, and read the output before you share it.
Mistakes That Leak Data Into AI Prompts
Pasting a stack trace that contains a customer’s email
An email address sits in the middle of a long trace, and you paste the whole thing into an AI chat without noticing it.
💡 See the solution▼
Trusting the output without reading it
The tool finds what its patterns describe. An ID in a custom format, a name, or a street address can pass through untouched.
💡 See the solution▼
Re-running the tool before restoring the AI’s reply
You change the input or a setting mid-conversation. The mapping is replaced, and tokens in earlier AI replies no longer point to the right values.
💡 See the solution▼
Sharing the mapping CSV along with the log
The CSV lists every token next to its real value. Sending it with the log undoes the redaction.
💡 See the solution▼
Expecting partial masks to be restorable
You pick Partial mask, then try to restore the AI’s reply. Different values can produce the same mask, so there is no safe way to reverse it.
💡 See the solution▼
Questions About Redacting Logs for AI
Q.What does the log anonymizer detect?
A. It detects IPv4 and IPv6 addresses, email addresses, UUIDs (the 8-4-4-4-12 hex format used for many session IDs), hostnames, secrets, and US phone numbers. Secrets include values after labels such as api_key, token, secret, and password, plus Bearer tokens, JWTs, AWS access key IDs, GitHub tokens, and keys that start with sk-. Phone numbers are off by default because number patterns often match harmless text. You can also add up to five custom regex patterns for internal IDs such as customer or order numbers.
Q.Are my logs uploaded anywhere?
A. No. Detection, replacement, and the mapping table all run in JavaScript in your browser. The text you paste is not sent to our servers or to any other service, and it is gone when you reload or close the page.
Q.Why replace values with tokens instead of deleting them?
A. A token keeps the structure of the log. If one IP address appears three times, it becomes IP_1 all three times, so an AI assistant can still see that the errors came from the same source. Blacking out every value would hide that pattern and make the log harder to debug.
Q.Can I restore the real values after the AI replies?
A. Yes, in token mode. After you run the tool, paste the AI’s reply into the restore box and the tokens are swapped back using the mapping table. The mapping is kept in memory only. If you run the tool again with different input or settings, the mapping is replaced, so finish the AI conversation first. Partial masks cannot be restored because different values can share the same mask.
Q.What is the difference between tokens and a partial mask?
A. Tokens replace each value with a numbered label such as IP_1 or EMAIL_1, and the same value always gets the same label. A partial mask keeps a recognizable piece of the value and hides the rest, for example 203.xxx.xxx.xxx or j*******@example.com. Tokens are better for AI debugging and are the only mode that supports restoring. Partial masks are easier to read at a glance.
Q.Will it catch every piece of sensitive data?
A. No. Detection is based on patterns, so it can miss unusual formats and can flag harmless text. For example, a version string like 1.2.3.4 or a browser string like Chrome/120.0.0.0 may be treated as an IP address. Hostnames are only detected when they end in a common domain suffix such as .com, .net, .io, .local, or .internal. Names and street addresses are not detected. Read the output before you share it.
Q.How do I hide a custom ID such as a customer or order number?
A. Open Custom patterns, enter a label, a JavaScript regular expression, and a token prefix. For example, the pattern CUST-\d{6} with the prefix CUSTOMER turns CUST-104233 into CUSTOMER_1. Custom patterns are case-sensitive and take priority over the built-in detectors when they overlap. A pattern that is not valid is flagged and skipped.
Q.Is the mapping table saved anywhere?
A. No. It lives in your browser’s memory while the page is open. If you download the CSV, that file contains your real values, so handle it like the original log.
Your feedback
makes this tool better.
Tell us which features you'd like, what should be fixed, or which new tool you'd like to see. An engineer reads every submission and uses it to guide development.