Data masking

Q: How do I mask a list of email addresses?

Add * to the end of the path, such as emails*. Values in a plain list are numbered, emails[0], emails[1], and the path emails on its own does not match them. Objects inside a list, such as orders[0].email, are matched by email as usual.

For: Account owners, administrators and data-protection reviewersRequires: Any signed-in user can view rules and test masking; changes need the Admin roleScreenshots show sample data

The data masking page holds the rules that hide parts of sensitive values in your API’s responses. When the assistant calls one of your tools, StrikeUp masks the response before the AI reads it and before it is stored in the tool-run log, so a card number reaches the AI as ************4242. Here you review the rules StrikeUp suggests, create your own, and test them on sample JSON. Open it from SecurityData masking; the page heading reads Privacy Data Masking.

What you can do here

Review the rules StrikeUp suggests. When a toolkit is generated from your API specification, StrikeUp proposes rules for fields that look sensitive. They mask nothing until you activate them. Activate suggested rules.
Create a rule for any field. Name the field, choose which toolkits and tools it covers, and how many characters stay visible at each end. Create a rule.
Check rules on sample data. Paste a JSON response and see the masked version the assistant would receive. Nothing is saved. Test masking.
Pause, change or delete a rule. Switch a rule off without losing it, or adjust what it keeps. Change, pause or delete a rule.
Find a rule by its reach. Filter the table by scope: every toolkit, one toolkit or one tool. Every control, explained.

The screen at a glance

The "Privacy Data Masking" card on the Data masking page. At the top right are "Refresh", "Test Masking" and the highlighted blue "New Rule" button, which opens the Create Masking Rule form. Below "Filter by Scope:" (set to All) the rules table lists seven rules with their field path, scope badge, toolkit and tool, kept characters, symbol, priority and status, for example "Card number" on payment_method.card_number (Global, 0 / 4, priority 95, Active), "VAT ID" for the Demo Billing API toolkit and "Billing contact phone", which is Inactive. — Every masking rule in one table; New Rule adds one and it applies from the next tool call.

Region	What it shows	What you do there
Suggested rules, at the very top	AI Suggested Masking Rules ({{count}}) with the field, name, toolkit and a preview for each suggestion. Shown only while suggestions are waiting	Activate or dismiss suggestions
Header	Privacy Data Masking, Refresh, Test Masking and New Rule	Reload rules, test, create a rule
Filter by Scope:	A scope filter	Narrow the table to one scope
Rules table	Every rule, active and inactive, highest priority first	Edit or delete a rule

How a rule masks a value

A rule keeps the first and last characters you choose and replaces every character in between with the masking symbol. The value keeps its length, so the AI can still tell a card number from a postcode.

Value	Keep first	Keep last	Result
`abcd`	2	2	`abcd`, unchanged: 4 characters is not longer than 2 + 2
`abcde`	2	2	`ab*de`
`jane.doe@example.com`	2	2	`ja****************om`
`4242424242424242`	0	4	`************4242`
`Robert`	3	3	`Robert`, unchanged: 6 characters is not longer than 3 + 3
`Roberta`	3	3	`Rob*rta`

A value is masked only when it is longer than the two numbers added together. Only text values are masked: a card number your API returns as a JSON number, such as 4242424242424242 without quotes, passes unchanged.

Which fields a rule matches

The field path is compared with each field’s position in the response, ignoring upper and lower case. Fields inside lists are numbered from 0, such as orders[0].customer.email.

Field path in the rule	Matches	Does not match
`email`	`email`, `user.email`, `orders[0].customer.email`	`emailVerified`
`user.email`	`user.email`, `account.user.email`	`user.emails[0]`
`*.iban`	`iban` and any field called `iban` at any depth	`iban_country`
`user`	every text value inside `user`, such as `user.name` and `user.phone`	`username`
`emails*`	`emails[0]`, `emails[1]`: a list of plain text values	—
`.account_number,.sort_code`	both fields; a comma separates several paths in one rule	—

A path that names an object masks everything inside it. user masks the whole user record, which is rarely what you want; name the field instead.

Which rule wins

For each field, StrikeUp takes the rules that apply to the toolkit and tool that answered, in this order, and uses the first one that matches:

Higher Priority first.
With equal priority, the narrower scope first: one tool, then one toolkit, then all toolkits.

Rule	Scope	Priority	Field path	Used for `billing_contact.email` from the tool listInvoices?
Billing contact email	One tool: listInvoices	60	`billing_contact.email`	Yes: highest priority that matches
Customer email	One toolkit	60	`*.email`	No: same priority, wider scope
Any email	Every toolkit	40	`email`	No: lower priority

Activate suggested rules

StrikeUp suggests rules when it generates a toolkit from your API specification. It reads the field names in the specification and proposes a rule for each field that looks like a password, key, card, bank, identity, contact or personal detail, with keep and priority values to match. Identifiers, counts, statuses and dates are not suggested.

Open SecurityData masking. Pending suggestions appear at the top in AI Suggested Masking Rules ({{count}}), grouped by kind, such as “Passwords & Secrets” and “Payment & Financial”. The group names are always in English.
Read each row: the field, the rule name with the reason, the toolkit, and Masked Preview.
Select Activate on a row, or Activate All for a whole group or every suggestion. The message {{count}} rule(s) activated confirms it, and the rule moves to the rules table as Active.
Select the ✕ icon on a row you do not want, or Dismiss All.

The "AI Suggested Masking Rules (2)" card at the top of the Data masking page, with "Activate All" and "Dismiss All" for every suggestion. Suggestions are grouped: under "Passwords & Secrets" the field payment_method.cvc, named "Card security code", for the Demo Billing API with masked preview "***"; under "Payment & Financial" the field refund.payout_iban, "Refund payout IBAN", with preview GB29…6819. Each row has its own "Activate" button and a dismiss icon. The highlighted "Activate" button in the card security code row turns that suggestion into an active rule. — Suggested rules mask nothing until you activate them; dismissed ones can be restored on the Suggestions page.

An activated suggestion becomes an ordinary rule for that toolkit, and you can edit it like any other. A dismissed suggestion stays switched off and is not suggested again for that toolkit; restore it from Suggestions. While suggestions wait, other pages show Masking rules suggested ({{count}}) — sensitive fields detected in your toolkits. with a Review them link.

The preview uses an invented sample value, so it can differ from the real result, for example for very short values. Use Test masking to see the real outcome.

Typical suggestions look like this:

Kind of field	Keep first	Keep last	Priority
Passwords, PINs, card security codes, secret keys	0	0	90
API keys and tokens	4	0	80
Card and bank account numbers, national ID and phone numbers	0	4	75
Email addresses, IBANs, passport numbers	2	4	65
Names, addresses, dates of birth	3	3	50
IP addresses	8	0	40

Create a rule

Select New Rule in the header.
Fill in Name * and Field Path *, for example *.email. Add a Description if colleagues need the reason.
Choose Scope *. For Toolkit-specific, also pick the toolkit; for Toolkit-Tool specific, the toolkit and the tool.
Set Keep First Characters, Keep Last Characters, Masking Symbol and Priority.
Select Create. The message Masking rule created successfully! appears and the dialog closes.

The "Create Masking Rule" dialog, filled in. "Name *" is "Customer email", the description reads "Hide most of every customer email address returned by the Accounts API.", "Field Path *" is *.email, "Scope *" is "Toolkit-specific" and "Toolkit *" is Demo Accounts API. "Keep First Characters" is 2, "Keep Last Characters" is 4, "Masking Symbol" is * and "Priority" is 60. The highlighted "Create" button at the bottom saves the rule as active straight away. — A filled-in rule that masks every email field in Accounts API responses, keeping the first 2 and last 4 characters.

A new rule is active at once and applies from the next tool call. Responses already stored in the tool-run log keep what they had.

Example. The rule in the screenshot masks every email field returned by the Demo Accounts API, keeping 2 characters at the start and 4 at the end: olivia.hart@example.com becomes ol*****************.com.

Test masking

Select Test Masking in the header.
Paste a sample response into JSON Data *. It must be valid JSON.
Optionally choose Toolkit (Optional) and Tool Name (Optional).
Select Test and compare Original JSON: with Masked JSON:.

The test uses the same rules and order as a live tool call:

Your choice	Rules used
No toolkit	Rules for all toolkits only
A toolkit	Rules for all toolkits and for that toolkit
A toolkit and a tool	Rules for all toolkits, for that toolkit and for that tool

The list under AI Suggested Masking Rules ({{count}}) is not involved: suggestions waiting for activation never mask. The applied-rules list names every active rule that applies to your choice, whether or not it changed a value, with its scope in StrikeUp’s internal wording (“Global”, “Server”, “ServerTool”). Compare the two JSON blocks to see what actually changed. Nothing is saved, and any member of your team can run tests.

Change, pause or delete a rule

Change a rule: select the pencil icon in its row, adjust the fields, and select Update. Scope, toolkit and tool cannot be changed; delete the rule and create a new one instead.
Pause a rule: in the same dialog, select Disable, then Update. The rule shows Inactive and masks nothing until you select Enable and save again.
Delete a rule: select the bin icon and confirm the browser’s question, Are you sure you want to delete this masking rule?. The rule is removed for good.

Why

Finds the rules for one toolkit in a long list.

Rules table columns

Column	What it shows	Why
Name	The rule name, with its description underneath	Tells colleagues what the rule protects
Field Path	The path or paths the rule matches	Check the path matches your API’s field names
Scope	Global (blue), Toolkit (purple) or Toolkit Tool (orange)	Shows how far the rule reaches
Toolkit/Tool	The toolkit, and the tool for single-tool rules; empty for rules on all toolkits	Shows which API the rule belongs to
Keep Chars	Characters kept at the start / at the end	Shows how much stays readable
Symbol	The masking character	—
Priority	0–100; the table is sorted by it, highest first	Decides which rule wins
Status	Active or Inactive	Inactive rules mask nothing
Actions	Pencil (edit) and bin (delete) icons	Change or remove the rule

Create Masking Rule dialog

Field	Default	Rules
Name *	Empty	Required. Missing name or path shows Name and Field Path are required
Description	Empty	Optional
Field Path *	Empty	Required. Dot notation, `*` wildcards, commas between several paths. The path is not checked against your API
Scope *	Global (all toolkits)	Or Toolkit-specific, or Toolkit-Tool specific
Toolkit *	—	Required for the two narrower scopes: Toolkit is required for toolkit-specific rules. Lists every toolkit, including switched-off ones
Tool Name *	—	Required for a single-tool rule: Tool name is required for toolkit-tool specific rules. Matches the tool’s exact name
Keep First Characters	2	0–50
Keep Last Characters	2	0–50
Masking Symbol	`*`	One character. Do not leave it empty
Priority	0	0–100; Higher priority rules are applied first

The Edit Masking Rule dialog has the same fields except scope, toolkit and tool, plus the Disable or Enable button.

Test Masking dialog

Where

Test Masking in the header.

What

JSON Data * takes the sample; Toolkit (Optional) and Tool Name (Optional) choose which rules apply. Empty input shows Please enter JSON data to test; input that is not JSON shows Invalid JSON format. Close clears the dialog.

Why

Shows the exact masked text the assistant would receive.

Who sees what

Part	Who can use it
The page, rules table and suggestions, read-only	Every signed-in member
Test Masking	Every signed-in member
Creating, editing, pausing and deleting rules; activating and dismissing suggestions	The Admin role only. Other roles see the buttons, but the change fails with Request failed

What you need to know

Topic	Fact
What is masked	Successful responses from your toolkits’ tools, before the AI reads them and before they are written to the tool-run log.
What is not masked	What customers type in the chat, the assistant’s replies, the request the assistant sends to your API (it is stored in the tool-run log as sent), knowledge base content, and messages on the violations page.
Error responses	When your API answers with an error, the AI receives the error text unmasked; only the copy in the tool-run log is masked. Keep personal data out of your API’s error messages.
If masking fails, the response is withheld	If masking a response fails, for example because a rule’s masking symbol is empty, neither the AI nor the log gets the response: both see “[Error] The API response was withheld because the data-masking rules could not be applied.” and the assistant cannot use that tool result. Check every rule has a symbol.
One-way	Masking changes the value in the response. StrikeUp keeps no original, and there is no way to unmask.
Takes effect at once	New and changed rules apply from the next tool call. Older tool-run entries are not masked again.
Text values only	Numbers, true/false and empty values are never masked.
Per workspace	Rules belong to your workspace. A rule for all toolkits covers every toolkit of your workspace.
Languages	Matching uses field names, so it works whatever language the values are in. Suggestion group names are shown in English.
Cost	Masking itself uses no AI. Suggested rules are created as part of generating a toolkit.

Frequently asked questions

General

Does the AI ever see the unmasked value?

Not for a successful tool response covered by an active rule. It sees an unmasked value only when no active rule matches the field; error responses from your API are masked like any other. When masking a response fails, the AI gets no response at all, only a note that it was withheld. Test your rules with Test Masking to close gaps.

Are my customers’ chat messages masked?

No. Masking applies only to the responses of your toolkits’ tools. What a customer types, and what the assistant writes back, is not masked. The guardrails on SecurityGuardrails check customer messages, but they allow or block; they do not mask.

Can I unmask data later?

No. A rule replaces the characters in the response itself, and StrikeUp keeps no copy of the original value. The AI, the chat and the tool-run log only ever have the masked value. To see a full value, look it up in your own system.

Rules

Why wasn’t a short value masked?

A value is masked only when it is longer than Keep First Characters and Keep Last Characters added together. With 3 and 3, “Robert” (6 characters) stays as it is and “Roberta” becomes “Rob*rta”. Lower the numbers for fields with short values.

Why wasn’t a number masked?

Only text values are masked. If your API returns a card number as a JSON number, without quotes, no rule changes it. Ask your API team to return such values as text, or remove them from the response.

Does the path email also mask customer.email?

Yes. A path matches the field with that name at any depth, so email covers email, customer.email and orders[0].customer.email. Upper and lower case are ignored. Write customer.email instead if you only want that one.

How do I mask a list of email addresses?

Add * to the end of the path, such as emails*. Values in a plain list are numbered, emails[0], emails[1], and the path emails on its own does not match them. Objects inside a list, such as orders[0].email, are matched by email as usual.

Two rules match the same field. Which one wins?

The one with the higher Priority. With equal priority, the narrower scope wins: a single-tool rule before a toolkit rule before a rule for all toolkits. Only one rule applies to each field, so a later rule never masks further.

Can I change a rule’s scope or toolkit?

No. The edit dialog changes name, description, path, the kept characters, symbol, priority and status only. To move a rule to another scope or toolkit, create the new rule with New Rule and delete the old one.

How do I switch a rule off without deleting it?

Select the pencil icon in the rule’s row, select Disable, then Update. The rule shows Inactive and stops masking from the next tool call; switch it back on the same way.

Suggested rules

Where do suggested rules come from?

StrikeUp creates them when it generates a toolkit from your API specification. It reads the field names in the specification and proposes rules for fields that look like secrets or personal data. The suggested rules belong to that toolkit and wait, switched off, until you activate them.

Do suggested rules mask anything before I activate them?

No. A suggestion masks nothing until you select Activate or Activate All on SecurityData masking. After that it is an ordinary active rule that you can edit or pause.

I dismissed a suggestion by mistake. Can I get it back?

Yes. Dismissed suggestions are kept, switched off. Restore them from ConfigureSuggestions, then activate them on the data masking page. A dismissed field is not suggested again on its own when the toolkit is regenerated.

Why does the masked preview differ from the real result?

The preview masks an invented sample value, not your data, and always shows at least a few masking characters. Real values that are too short stay unmasked. Paste a real-looking response into Test Masking to see the actual result.

Testing

The test lists a rule that changed nothing. Why?

The applied-rules list in Test Masking names every active rule that applies to the toolkit and tool you chose, whether or not its path matched a field. Compare the original and masked JSON to see which values actually changed.

Can colleagues without the Admin role test masking?

Yes. Every signed-in member can open Test Masking and run tests; nothing is saved. Creating, editing, deleting, activating and dismissing rules need the Admin role.

Data and privacy

Is the request the assistant sends to my API masked in the log?

No. The tool-run log stores the request exactly as sent, including any values the customer gave the assistant. Only your API’s response is masked. Keep this in mind for tools that take personal data as input.

Do new rules mask older tool-run entries?

No. A rule applies from the next tool call. Entries already in the tool-run log keep the values they were stored with, masked or not.

Troubleshooting

Symptom	What to check
A value still appears in full in a tool run	In order: is the rule Active? Does the path match the field (try Test Masking)? Is the value text, not a number? Is it longer than the kept characters together? Do the toolkit and the exact tool name match? Does a higher-priority rule keep more characters?
Nothing is masked any more for a toolkit	Look for a rule with an empty Symbol column. Give it a symbol, such as `*`, and save.
Name and Field Path are required	Fill in both the name and the field path.
Toolkit is required for toolkit-specific rules or Tool name is required for toolkit-tool specific rules	Pick the toolkit, and for a single-tool rule the tool, or choose a wider scope.
Invalid JSON format	The sample must be valid JSON: double quotes around names and text, no trailing commas.
Failed to activate masking rule, Failed to dismiss masking rule or Request failed	Changes need the Admin role.
The suggestions card is missing	No suggestions are waiting. Dismissed ones are on ConfigureSuggestions. Reload the page if you expected new ones.
The preview and the real result differ	The preview uses a sample value. Use Test Masking with a realistic response.
A whole object is masked	The path names the object, such as `user`. Name the field instead, such as `user.email`.