Replace values, preserve the sentence.
A typed placeholder gives the model a reference it can carry into a response. The surrounding task remains part of the prompt your application sends.
AI APPLICATIONS & WORKFLOWS
Give the model context.
Tokenize personal details.
Replace identified sensitive values with placeholders before a model call. Your application uses Databunker Pro to store the mapping and resolve permitted values afterward.
DETECT → TOKENIZE → CALL → RESOLVE
The application orchestrates the flow.
Detection happens before tokenization. The model receives the prompt your application constructs and never connects directly to the vault.
Separate detection step
Your detector
Identifies the value and its location in the input.
Email detected[email protected]
Your application
Build the prompt
Draft a reply to [email protected] about their delivery.
Credential: authorized vault callsModel provider
Work with context
Receives the tokenized prompt. Has no direct vault connection.
WAITING FOR APPLICATIONDatabunker Pro
Encrypted token mapping
Created by the application’s token API call
TOKEN REFERENCE
[EMAIL:a21f…]BACKING STORAGEEncrypted mapping in the vault database
ILLUSTRATIVE RESPONSE
The response appears after the model call.The detector identifies values. Databunker stores token mappings. Your application makes both the model call and the authorized resolution call.
01INTERACTIVE EXAMPLE
Identify values before tokenization.
A separate detector identifies sensitive values in the text. This illustrative detector finds the email address; Databunker does not make that detection step automatic.
A local demonstration using fictional data. The model reply is illustrative; no detector, model, or vault API is called.
Identified original valueTyped placeholderApplication-held credential
Control how originals are resolved.
Use the vault’s token APIs and scoped credentials to retrieve the values needed by the application. The record reference is not the access credential.
Choose the token lifecycle.
Configure expiry and choose whether repeated values should reuse a token. Decide which workflow needs continuity and how long its mappings should remain available.
BUILD THE FIRST WORKFLOW
Start with one prompt path.
Choose an application flow with known sensitive fields. Test the complete path, including the response and any traces your application records.
Read the tokenization guide ↗- Identify the values.For free text, use your detector to select what must be replaced.
- Create and substitute tokens.Store the values through the token API and construct the prompt placeholders.
- Call the model from your application.Keep vault credentials outside the model’s prompt and tools.
- Resolve authorized values.Restore the values required by the workflow and handle the resulting response deliberately.
Explore another workflow
See the database migration Move the source profiles into the vault.
LESS SPRAWL. MORE CONTROL.
Give sensitive data
a proper home.
Walk through your architecture with a Databunker engineer.
Find the right place to start.