github Download

Tokenization #

Tokenization (also known as tokenisation) is a way to transform some sensitive data into pseudo-random representation of same data type.

A number (32 or 64-bit signed integer) will remain a number. String will be string. There is also special variant for string that contains email, it will be transformed to email-looking string as well.

It is better to see than to read:

  • 789156749362
  • testing testinglwcIvM3zaJN9cbN
  • kate@example.comsVAOfOwF@mSuGPog.et

The tokenization process has few interesting properties:

  • Tokenized value is simply random, it is neither encrypted nor hashed version of the input;
  • Tokenization supports two modes:
    • Consistent tokenization — result will always remain the same for the same input (hello might be tokenized into wshfwSjdsn),
    • Inconsistent tokenization — result will be different every time, even for the same input (hello might be tokenized into KishJs or KdCbshQoP or KeitAyheof),
  • Tokenization can be reversed, but only for valid (previously returned) values.

Two components can provide tokenization functionality:

  • AcraServer — transparent tokenization for INSERT and UPDATE queries, transparent detokenization for SELECT queries, with per column configuration.
  • AcraTranslator — provides gRPC and HTTP API for tokenization.

Consistent tokenization might be applied similarly to searchable encryption to search tokenized data in DB.

Under the hood, if the field marked with the consistent_tokenization option AcraServer will detect all queries with WHERE equal clauses (all types of statements could be used SELECT/UPDATE/INSERT/DELETE) and transform them to apply to search.

AcraServer extract right expression in WHERE condition and replace it with tokenized data.

This means queries like:

SELECT/UPDATE/INSERT/DELETE ... FROM ... WHERE consistent_tokenized_column = "value"

eventually in DB, will look like:

SELECT/UPDATE/INSERT/DELETE ... FROM ... WHERE consistent_tokenized_column = <consistent_tokenized_value>
Currently, AcraTranslator only supports consistent tokenization. If you need inconsistent tokenization, contact us to get this feature in Acra Enterprise Edition.

Both tokenization types require deploying an additional database, Redis by default, to store pairs: token <-> encrypted data.

Refer to Acra in depth / Architecture and Configuring and maintaining / Key storing to learn more about Redis token storage.

AcraServer configuration #

In configuration file, passed by --encryptor_config_file flag, you can individually configure tokenization for any field of any table, if that field has tokenizable type.

These options are accepted:

  • tokenized — boolean, use true to enable tokenization.

    Deprecated since 0.93.0: just specify token_type instead, and it will automatically enable the tokenization.

  • consistent_tokenization — boolean, use true to make tokenization consistent

  • token_type — string representing type of token:

    • int32 — 32-bit signed integer
    • int64 — 64-bit signed integer
    • str — text string
    • bytes — byte string (for column types like bytea, BLOB, VARBINARY)
    • email — text string that contains email, tokenized version will also look like email
    • int32_string and int64_string — similar to int32 and int64, but for columns that contain number in text (decimal) representation, that number will be parsed, tokenized, and then converted to text again

AcraTranslator API #

Among all the crypto-related operations, AcraTranslator includes tokenization. Both protocols (gRPC and HTTP) provide the same functionality, but from different angles. You decide which one fits best for your application.

Zones are deprecated since 0.94.0, will be removed in 0.95.0. ZoneID parameters will be removed or ignored. AcraTranslator will expect explicitly specified ClientID in gRPC API or will use ClientID from TLS certificates.

gRPC #

In order to use the translator gRPC API you have to take api.proto, and use either service Tokenizator or service BulkProcessing (with one or more TokenizeRequest inside, enterprise only).

Tokenize & detokenize with service Tokenizator:

message TokenizeRequest {
    bytes client_id = 1;
    bytes zone_id = 2;
    oneof value {
        string str_value = 3;
        string email_value = 4;
        int32 int32_value = 5;
        int64 int64_value = 6;
        bytes bytes_value = 7;

message TokenizeResponse {
    oneof response {
        string str_token = 1;
        string email_token = 2;
        int32 int32_token = 3;
        int64 int64_token = 4;
        bytes bytes_token = 5;

service Tokenizator {
    rpc Tokenize (TokenizeRequest) returns (TokenizeResponse) {}
    rpc Detokenize (TokenizeRequest) returns (TokenizeResponse) {}
Detokenization is performed by creating TokenizeRequest with the token as value and passing it to Detokenize method. Don’t forget to use the same type (str / email / int32 / int64 / bytes).


Request #

Method: POST (available since 0.91.0), GET (deprecated since 0.91.0)

Path: /v2/tokenize for tokenization, /v2/detokenize for detokenization

Mime-Type: application/json


{ "zone_id": "",
  "type": 3,
  "data": "hidden message" }

zone_id is optional field specifying the zone ID you want to associate with this request

type is a number that tells AcraTranslator the type of value being tokenized

  • 1 — 32-bit signed integer
  • 2 — 64-bit signed integer
  • 3 — text string
  • 4 — byte string (for column types like bytea, BLOB, VARBINARY)
  • 5 — text string that contains email, tokenized version will also look like email

data contains data (for tokenization) or token (for reverse operation), the exact type depends on type: for integral types you’d make data a number, for string or email — string, and for bytes — base64-encoded string

Response #

Status code: 200

Mime-Type: application/json


{ "data": "CWce2KYUPVYHlq" }

type of data will depend on request, no matter what operation you perform, you will get same looking thing, string for string, number for number and so on

Examples #

You can use this docker-compose file as a playground, it will bring up AcraTranslator and expose its HTTP server at

Here you can see few examples using curl. HTTP API allows bulk requests as well as simple ones (one at a time). This example includes few simple ones.

curl \
    --request POST \
    --header 'Content-Type: application/json' \
    --data '{"zone_id":"","type":3,"data":"hidden message"}' \

curl \
    --request POST \
    --header 'Content-Type: application/json' \
    --data '{"zone_id":"","type":5,"data":"me@domain.com"}' \

Here type can be a number from 1 to 5, where 1 is int32 and 5 is email (see list of possible values for token_type in AcraServer configuration, the order is the same).