Skip to main content
Integrating the SDK into your app is easy, and its footprint (memory and network usage) is low. Getting started: For help getting started, see the developer guide. Latest version: 0.8.5. See the C++ SDK changelog for release notes, or the SDK changelog index for all SDKs. SDK methods: For the full reference documentation of the C++ SDK, see the reference section.

Developer guide

This guide is designed to help you integrate the C++ SDK quickly and start evaluating feature flags in your C++ app.

Getting started

Install the C++ client

The C++ SDK is a C++17 library built on top of the Kameleoon SDK Core, a native shared library (libkameleoon_ffi) written in Rust that implements the SDK behavior. Your app links the C++ library statically and loads the core library at runtime.
The Kameleoon SDK Core is compiled from source for your target platform and architecture. Make sure Rust and Cargo are installed and available in your system’s PATH, and install cbindgen, which generates the C header of the core library:
Install Rust and Cargo: https://rust-lang.org/tools/installFuture releases will provide prebuilt core libraries for common platforms to simplify installation. Local compilation will remain available for users who prefer to build from source.
Requirements:
  • A C++17 compiler (GCC, Clang, or MSVC) and CMake 3.16 or later.
  • Rust 1.91 or later, Cargo, and cbindgen to build the SDK Core library (libkameleoon_ffi.so on Linux, libkameleoon_ffi.dylib on macOS, kameleoon_ffi.dll with its import library kameleoon_ffi.dll.lib on Windows).
Clone the C++ SDK repository and build the core library, then build the C++ library against it:
To integrate the SDK into your own CMake project, add the cpp/client directory with add_subdirectory() and link the Kameleoon::CppClient target. Point KAMELEOON_FFI_LIBRARY to the core library you built; the C header is generated with cbindgen, or you can pass an existing one with KAMELEOON_FFI_HEADER:
CMakeLists.txt
  • libkameleoon_ffi is loaded at runtime: install it to a standard library location, or ship it next to your executable. On Linux and macOS, shipping it next to the executable requires an RPATH of $ORIGIN or @loader_path (for example, set_target_properties(my_app PROPERTIES INSTALL_RPATH "$ORIGIN")). On Windows, copy kameleoon_ffi.dll next to your binaries.
  • When you create a client, the SDK checks that the loaded libkameleoon_ffi matches the version it was built against and throws a KameleoonException with ErrorCode::Internal otherwise. Always upgrade the C++ library and the core library together.
  • All public SDK types live in the kameleoon namespace. Include kameleoon/kameleoon.hpp to get the whole public API.

Additional configuration

Create a JSON configuration file (for example, client-cpp.json) to provide credentials and customize SDK behavior. You can also download a sample configuration file. The C++ SDK can be configured either with a JSON file whose path you pass to KameleoonClientFactory::create(), or by filling a KameleoonClientConfig struct directly in code. The following table shows the available properties that you can set:

Initialize the Kameleoon client

After you have installed the SDK and configured your credentials, create a KameleoonClient by using KameleoonClientFactory.
The examples on this page use C++20 designated initializers (GetVariationOptions{.track = false}) to fill the SDK’s configuration, options, and data structs. With a C++17 compiler, declare the struct and set its fields one by one instead.
A KameleoonClient is the main object used to evaluate feature flags, add visitor data, and send tracking requests.
  • It’s recommended to use KameleoonClient as a singleton object, as it serves as the bridge between your app and the Kameleoon platform. It exposes all required methods and properties to run experiments efficiently. KameleoonClient is move-only: keep it in one place and pass it by reference.
  • The C++ SDK initializes asynchronously. You should call initialize() before relying on feature evaluation in production code.
  • Every SDK failure is reported as a KameleoonException. Wrap SDK calls in try/catch blocks.

Activating a feature flag

Assigning a unique ID to a user
To assign a unique ID to a user, you can use the get_visitor_code() method. If a visitor code doesn’t exist (from the request headers cookie), the method generates a random unique ID or uses a default_visitor_code that you would have generated. The ID is then set in a response headers cookie. If you are using Kameleoon in Hybrid mode, calling the get_visitor_code() method ensures that the app file engine.js (previously named, kameleoon.js) and the SDK share the unique ID (visitor code).
Retrieving a flag configuration
To implement a feature flag in your code, you must first create the feature flag in your Kameleoon account. To determine the status or variation of a feature flag for a specific user, you should use the get_variation() or is_feature_active() method to retrieve the configuration based on the feature_key. The get_variation() method handles both simple feature flags with ON/OFF states and more complex flags with multiple variations. The method retrieves the appropriate variation for the user by checking the feature rules, assigning the variation, and returning it based on the feature_key and visitor_code. You can use the is_feature_active() method to retrieve the configuration of a simple feature flag that has only an ON or OFF state, as opposed to more complex feature flags with multiple variations or targeting options. If your feature flag has associated variables (such as specific behaviors tied to each variation) get_variation() also enables you to access the Variation object, which provides details about the assigned variation and its associated experiment. This method checks whether the feature flag targets the user, finds the visitor’s assigned variation, and saves it to storage. When track=true, the SDK sends the exposure event to the specified experiment on the next tracking request, which it triggers automatically based on the SDK’s tracking_interval_millis. The default interval is 1000 milliseconds (1 second). The get_variation() method allows you to control whether the SDK tracks the visitor. If track=false, the SDK sends no exposure events. This option is useful if you prefer not to track data through the SDK and instead rely on client-side tracking managed by the Kameleoon engine, for example. Additionally, setting track=false is helpful when using the get_variations() method, where you might only need the variations for all flags without triggering any tracking events. If you want to know more about how tracking works, view this article
Adding data points to target a user or filter / breakdown visits in reports
To target a user, ensure you’ve added relevant data points to their profile before retrieving the feature variation or checking if the flag is active. Use the add_data() method to add these data points to the user’s profile. To retrieve data points collected on other devices or to access past user data (collected client-side when using Kameleoon in Hybrid mode), use the get_remote_visitor_data() method. This method asynchronously fetches data from the servers. It’s important to call get_remote_visitor_data() before retrieving the variation or checking if the feature flag is active, because the SDK might need this data to assign a user to a given variation. To learn more about available targeting conditions, see the detailed article on the subject. Additionally, the data points you add to the visitor profile will be available when analyzing your experiments, allowing you to filter and break down your results by factors like device and browser. Kameleoon Hybrid mode automatically collects a variety of data points on the client-side, making it easy to break down your results based on these pre-collected data points. See the complete list here. If you need to track additional data points beyond what’s automatically collected, you can use Kameleoon’s Custom Data feature. Custom Data allows you to capture and analyze specific information relevant to your experiments. Don’t forget to call the flush() method to send the collected data to Kameleoon servers for analysis.
To ensure your results are accurate, it’s recommended to filter out bots by using the UserAgent data type.
Tracking goal conversions
When a user completes a desired action (such as making a purchase), it’s recorded as a conversion. To track conversions, use the track_conversion() method and provide the required visitor_code and goal_id parameters. The SDK sends the conversion tracking request along with the next scheduled tracking request, which occurs at regular intervals (defined by tracking_interval_millis). If you prefer to send the request immediately, use the flush_instant() method.
Sending events to analytics solutions
To track conversions and send exposure events to your customer analytics solution, you must first implement Kameleoon in Hybrid mode. Then, use the get_engine_tracking_code() method. The get_engine_tracking_code() method retrieves the unique tracking code required to send exposure events to your analytics solution. Using this method allows you to record events and send them to your desired analytics platform.

Error handling

Every error the SDK raises is a KameleoonException, so a single handler for that type catches every SDK failure. Handle a specific error type when your app needs to react to a particular failure, for example to fall back to your default behavior when a feature flag isn’t in the configuration or the SDK isn’t initialized yet. The reference section of each method lists the errors it can raise. KameleoonException derives from std::runtime_error. Call code() to get the ErrorCode that tells the errors apart, and what() to get the message.
Strings passed to the SDK must be valid UTF-8. Callbacks you provide to the SDK (CookieAccessor overrides, event handlers, and loggers) must not throw: any exception that escapes them is caught at the SDK boundary and discarded.
SDK errors are expected behavior, not crashes: a feature flag that isn’t activated yet, an SDK that is still loading its configuration, or a network failure are all normal conditions. When feature evaluation fails, log the error and fall back to your default behavior so the visitor still gets a working experience.

Asynchronous operations

The network-bound operations (initialize(), flush_instant(), get_remote_data(), get_remote_visitor_data(), and get_visitor_warehouse_audience()) run on the SDK’s own worker thread. Each comes in three forms, so you can pick the one that fits the calling thread:
A completion is a Completion<void> (std::function<void(std::exception_ptr)>), or a Completion<T> (std::function<void(std::exception_ptr, T)>) for operations that return a value. The SDK invokes it exactly once. You can omit the completion to run the operation fire-and-forget: failures then only appear in the SDK log.
Completions and event handlers usually run on the SDK worker thread, but a completion can also run on the calling thread, before the *_async call returns: for example, when the SDK fails fast (the client isn’t ready, the visitor code is invalid) or when initialize_async() is called on a client that is already initialized. Don’t assume a completion runs on a particular thread or after the call returns, and don’t hold a lock across the call that the completion also takes. Completions and event handlers must not block, must not throw, and must not call the blocking form of an SDK operation: it waits for the same worker thread and deadlocks. Hand the outcome to your own event loop and finish the work there.
C++20 has no built-in coroutine task type, so co_await only works inside a coroutine whose promise type accepts the SDK’s Awaitable<T> (KAMELEOON_HAS_COROUTINES is defined by kameleoon/async.hpp when coroutines are available). By default the coroutine resumes on the SDK worker thread; call .via(executor) on the awaitable to resume on your own event loop instead. Frameworks with their own coroutine types, such as Boost.Asio, need a small adapter around the completion form; see the server example shipped with the SDK.
The evaluation methods (get_visitor_code(), get_variation(), get_variations(), is_feature_active(), add_data(), track_conversion(), and flush()) are synchronous in-memory calls and need none of this.

Using a custom bucketing key

By default, Kameleoon uses a unique, anonymous visitor ID (visitor_code) to assign users to feature flag variations. This ID is typically generated and stored on the user’s device (in a browser cookie for client-side and server-side SDKs, and in persistent storage for mobile SDKs). However, in certain scenarios you may need to ensure all users of the same organization see the same variant of a feature flag. The Custom Bucketing Key option allows you to override this default behavior by providing your own custom identifier for bucketing. This override ensures that Kameleoon’s assignment logic uses your specified key instead of the default visitor_code.

Use cases

Using a custom bucketing key is essential for maintaining consistency and accuracy in your feature flag assignments, particularly in these situations:
  • Account-level or organizational experiments: For B2B products or scenarios where you want to assign all users from the same organization to the same variation, you can use an identifier like an account_id. Custom bucketing keys are crucial for A/B testing features that impact an entire team or company.
By implementing a custom bucketing key, you ensure greater consistency and accuracy in your experiments, which leads to more reliable results and a better user experience.

Technical details

When you configure a custom bucketing key for a feature flag, you provide Kameleoon with a specific identifier from your app’s data:
  • Providing the custom key: You provide your custom identifier to the Kameleoon SDK using the add_data() method. In this method, you will pass your chosen custom bucketing key as a CustomData object. Here, new_visitor_code refers to the identifier you wish to use for your bucketing (for example, the new user_id or account_id).
For the custom bucketing key to function correctly, you must also define and configure it for the feature flag during the flag creation or editing process. Without this corresponding configuration, the SDK’s bucketing won’t apply your custom key. For detailed instructions on how to set this up in Kameleoon, refer to this article.
  • Bucketing logic: Once you provide a custom bucketing key through the add_data() method, all hash calculations for assigning users to variations use this new_visitor_code (your custom key) instead of the default visitor_code. Using the new_visitor_code means that the bucketing decision depends on your custom identifier, ensuring consistent assignments across various contexts where that identifier is present.
  • Data tracking and analytics: It’s crucial to note that while the new_visitor_code (your custom key) is used for bucketing decisions, the SDK sends all subsequent data (tracking events and conversions, for example) and associates it with the original visitor_code. This separation ensures that your analytics accurately reflect individual user journeys and interactions within your experiment’s broader context, even when you bucket at a higher level (like an account) or across multiple devices/sessions. Your original visitor data remains intact for comprehensive reporting.

Technical requirements

To effectively use a custom bucketing key:
  • The key must be a std::string.
  • It must be unique for the entity you intend to bucket (for example, if using a user_id, each user’s ID should be unique).
  • The key must be available to the SDK at the exact moment it evaluates the feature flag decision for that user or request.

Targeting conditions

The Kameleoon SDKs support a variety of predefined targeting conditions that you can use to target users in your campaigns. For the list of conditions this SDK supports, see use visit history to target users. You can also use your own external data to target users.

Cross-device experimentation

To support visitors who access an app from multiple devices, Kameleoon allows the synchronization of previously collected visitor data across each of the visitor’s devices and reconciliation of their visit history across devices through cross-device experimentation. Case studies and detailed information on how Kameleoon handles data across devices are available in the article on cross-device experimentation.

Synchronizing custom data across devices

Although custom mapping synchronization aligns visitor data across devices, it’s not always necessary. Below are two scenarios where custom mapping sync isn’t required: Same user ID across devices If you use the same user ID consistently across all devices, the SDK handles synchronization automatically without a custom mapping sync. It’s enough to call the get_remote_visitor_data() method when you want to sync the data collected between multiple devices. Multi-server instances with consistent IDs In complex setups involving multiple servers (for example, distributed server instances), where the same user ID is available across servers, synchronization between servers (with get_remote_visitor_data()) is sufficient without additional custom mapping sync. Customers who need additional data can refer to the get_remote_visitor_data() method description for further guidance. In the below code, it’s assumed that you use the same unique identifier (in this case, the visitor_code, also known as userId) consistently between the two devices for accurate data retrieval.
If you want to sync collected data in real time, you need to choose the scope Visitor for your custom data.
Device A
Device B

Using custom data for session merging

Cross-device experimentation allows for combining a visitor’s history across each of their devices (history reconciliation). History reconciliation allows merging different visitor sessions into one. To reconcile visit history, use CustomData to provide a unique identifier for the visitor. For more information, see the dedicated documentation. After you enable cross-device reconciliation, calling get_remote_visitor_data() with the parameter userId retrieves all known data for a given user. Sessions with the same identifier always see the same variation in an experiment. In the Visitor view of your experiment’s results pages, these sessions will appear as a single visitor. The SDK configuration ensures that associated sessions always see the same variation of the experiment. However, there are some limitations regarding cross-device variation allocation. The cross-device experimentation documentation outlines these limitations. Follow the activating cross-device history reconciliation guide to set up your custom data on the Kameleoon platform. Afterwards, you can use the SDK normally. The following methods that may be helpful in the context of session merging:
  • get_remote_visitor_data() with added UniqueIdentifier(true) - to retrieve data for all linked visitors.
  • track_conversion() or flush() with added UniqueIdentifier(true) data - to track some data for specific visitor that’s associated with another visitor.
Because the custom data you use as the identifier requires Visitor scope, you need to use cross-device custom data synchronization to retrieve the identifier with the get_remote_visitor_data() method on each device.
Here’s an example of how to use custom data for session merging.
In this example, the app has a login page. Since the user ID is unknown at the moment of login, the app uses an anonymous visitor identifier generated by the get_visitor_code() method. After the user logs in, the app associates the anonymous visitor with the user ID and uses it as a unique identifier for the visitor.

Logging

The SDK generates logs to reflect various internal processes and issues.

Log levels

The SDK supports configuring limiting logging by a log level.

Custom handling of logs

The SDK writes its logs to the console output by default. This behaviour can be overridden.
The SDK limits logging by log level separately from the log handling logic.

Reference

This is the full reference documentation for the C++ SDK.

Initialization

create()

To use the SDK, create a KameleoonClient with KameleoonClientFactory::create(), either from a KameleoonClientConfig struct or from a JSON configuration file. Clients are cached per site code and environment: repeated create() calls with the same site code and environment return clients over the same underlying SDK instance, and the configuration of the first call wins.
Parameters
Return value
Exceptions

initialize()

Use initialize() when your app should wait for the Kameleoon client to finish initialization before it evaluates feature flags. The call returns the result of the configuration fetch: it completes successfully when the client finishes initializing, and fails if the initial configuration fetch fails before the client becomes ready. Background retries continue after that first failure. Once a retry succeeds, subsequent calls return a successful result. If no initialization result is available before the timeout expires, the call fails. If you don’t provide a timeout value, the SDK uses the default timeout from default_timeout_millis.
Call initialize() once on app startup to wait until the SDK has loaded its configuration or the timeout expires. On each incoming request, use is_ready() as a non-blocking guard before evaluating feature flags. Even if initialize() fails, the SDK keeps retrying the configuration fetch in the background, so is_ready() starts returning true after a retry succeeds.
Parameters
Return value
Exceptions

is_ready()

is_ready() checks whether the SDK is ready for use, which means its configuration has been successfully loaded. Unlike initialize(), this method returns immediately without blocking or throwing.
  • Once the SDK becomes ready, it stays ready. If a later configuration refresh fails, the SDK keeps using the previously downloaded configuration and doesn’t return to a not-ready state.
  • Call initialize() once on app startup to wait until the SDK has loaded its configuration or the timeout expires. On each incoming request, use is_ready() as a non-blocking guard before evaluating feature flags. Even if initialize() fails, the SDK keeps retrying the configuration fetch in the background, so is_ready() starts returning true after a retry succeeds.
Return value

forget()

Removes the cached SDK client associated with the specified site_code, so the next create() call with the same site code and environment creates a new one. forget() only drops the factory’s reference. Every KameleoonClient created for that site code keeps the underlying SDK instance (and its worker thread, polling, and tracking) alive until it’s destroyed, and outstanding asynchronous operations keep it alive until their completions finish. To release the background resources, call forget() and destroy all KameleoonClient instances of that site code.
Parameters

Feature flags and variations

is_feature_active()

  • 📨 Sends tracking data to Kameleoon (depending on the track option)
Determines whether a feature flag is active for a given user. If the visitor hasn’t yet been evaluated for this feature flag, the SDK evaluates the targeting rules and returns the result. If the visitor already has a stored evaluation for the feature, the SDK reuses the existing result to ensure consistency.
Kameleoon uses tracking to count sessions and visitors when you call certain methods, such as is_feature_active(), get_variation() or get_variations().Use the default true value for the track parameter when you expose visitors to a variation and need to count them. Set the track parameter to false only if you call these methods before you expose visitors.For example, if you call get_variations() to retrieve all variations before you expose visitors, set the track parameter to false. This setting prevents Kameleoon from prematurely counting a session. You can then trigger tracking later when you explicitly expose the visitor.Kameleoon sends tracking data every second by default. You can configure this interval up to 5 seconds using the tracking interval configuration option. Kameleoon groups tracking events into a single session as long as the interval between events is less than 30 minutes. If more than 30 minutes elapse between tracking events, Kameleoon counts the events as separate sessions. A visit appears in your reports 30 minutes after the last recorded event in the session.
The is_feature_active() method evaluates the served variant, not the master flag state. If you exclude rules, the method uses the Then, for everyone else serve default state. If you select Off for this default state, the method always returns false even when the master feature flag is On.
Parameters
Return value
Exceptions

get_variation()

  • 📨 Sends Tracking Data to Kameleoon (depending on the track parameter)
Retrieves the Variation assigned to a given visitor for a specific feature flag. This method takes a visitor_code and feature_key as mandatory arguments. The track argument is optional and defaults to true. It returns the assigned Variation for the visitor. If the visitor isn’t associated with any feature flag rules, the method returns the default Variation for the given feature flag. Ensure your code includes proper error handling to manage potential exceptions.
The default variation refers to the variation assigned to a visitor when they don’t match any predefined delivery rules for a feature flag. In other words, it’s the fallback variation applied to all users who aren’t targeted by specific rules. It’s represented as the variation in the “Then, for everyone else…” section in a management interface.
Parameters
The optional track parameter is passed as the track field of a GetVariationOptions struct.
Return value
Exceptions

get_variations()

  • 📨 Sends Tracking Data to Kameleoon (depending on the track parameter)
Retrieves a map of Variation objects assigned to a given visitor across all feature flags. This method iterates over all available feature flags and returns the assigned Variation for each flag associated with the specified visitor. It takes visitor_code as a mandatory argument, while only_active and track are optional.
  • If you set only_active to true, the method get_variations() returns feature flag variations only if the visitor isn’t bucketed with the off variation.
  • The track parameter controls whether the method tracks the variation assignments. By default, it’s true. If it’s false, the method doesn’t track variation assignments.
The returned map consists of feature flag keys as keys and their corresponding Variation as values. If the SDK assigns no variation for a feature flag, the method returns the default Variation for that flag. Implement proper error handling to manage potential exceptions.
The default variation refers to the variation assigned to a visitor when they don’t match any predefined delivery rules for a feature flag. In other words, it’s the fallback variation applied to all users who aren’t targeted by specific rules. It’s represented as the variation in the “Then, for everyone else…” section in a management interface.
Parameters
The optional only_active and track parameters are passed as the fields of a GetVariationsOptions struct.
Return value
Exceptions

set_forced_variation()

The method allows you to programmatically assign a specific Variation to a user, bypassing the standard evaluation process. The method is especially valuable for controlled experiments where the usual evaluation logic isn’t required or you need to skip it. It can also be helpful in scenarios like debugging or custom testing. When you set a forced variation, it overrides Kameleoon’s real-time evaluation logic. The SDK skips processes like segmentation, targeting conditions, and algorithmic calculations. To preserve segmentation and targeting conditions during an experiment, set force_targeting=false instead.
Simulated variations always take precedence in the execution order. If a simulated variation calculation starts, the SDK fully processes and completes it first.
Kameleoon treats a forced variation the same as an evaluated variation. It’s tracked in analytics and stored in the user context like any standard evaluated variation, ensuring consistency in reporting. The method may throw exceptions under certain conditions (for example, invalid parameters, user context, or internal issues). Proper exception handling is essential to ensure that your app remains stable and resilient.
It’s important to distinguish forced variations from simulated variations:
  • Forced variations: Are specific to an individual experiment.
  • Simulated variations: Affect the overall feature flag result.
Parameters
The optional force_targeting parameter is passed as the force_targeting field of a SetForcedVariationOptions struct.
Exceptions

evaluate_audiences()

  • 📨 Sends Tracking Data to Kameleoon
This method evaluates visitors against all available Audiences Explorer segments and tracks those who match. Call evaluate_audiences() after you set or update all relevant visitor data, and just before getting a feature variation or checking a feature flag. This approach ensures that Kameleoon evaluates the visitor against the most current data available, allowing for accurate audience assignment based on all criteria. After calling this method, you can perform a detailed analysis of segment performance in Audiences Explorer.
Parameters
Exceptions
In most cases, you only need to handle the basic error, KameleoonException, as the example demonstrates. However, if different types of errors require a response, handle each one separately based on specific requirements. Additionally, for enhanced reliability, you can handle general language errors by including std::exception.

get_datafile()

To evaluate all feature flags, use get_variations(). This method is more efficient than calling DataFile and iterating through flags with get_variation().
Returns the current SDK configuration as a DataFile object. The returned DataFile is a copy of the current configuration: it doesn’t change when the SDK refreshes its configuration.
Return value
Exceptions

Visitor data

get_visitor_code()

Use get_visitor_code() to obtain the current visitor’s Kameleoon visitor_code. The method works with any cookie store that implements the CookieAccessor interface. The implementation logic is as follows:
  1. The SDK checks whether a kameleoonVisitorCode cookie is already available through the provided accessor.
  2. If the cookie is absent, the SDK uses default_visitor_code when you provide one.
  3. Otherwise, the SDK generates a new visitor code and stores it through the accessor.
For more information, refer to Hybrid experimentation.
If you provide your own visitor_code, its uniqueness must be guaranteed on your side. Also note that the length of visitor_code is limited to 255 characters.
The get_visitor_code() method allows you to set simulated variations for a visitor. When cookies (from a request or document) contain the key kameleoonSimulationFFData, the method reads the cookie and stores the simulated variations for that visitor; its return value doesn’t change. Subsequent feature evaluations for the visitor, such as get_variation() and is_feature_active(), then bypass the standard evaluation process and return the simulated Variation based on the provided data.You can apply simulations in two ways:
  • Automatically (recommended): If using Kameleoon Web Experimentation or the SDK in Hybrid mode, Kameleoon creates the cookie automatically when you simulate a variant’s display using the Simulation Panel.
  • Manually: Set the kameleoonSimulationFFData cookie manually.
It’s important to distinguish simulated variations from forced variations:
  • Simulated variations: Affect the overall feature flag result.
  • Forced variations: Are specific to an individual experiment.
⚙️ Manual setupEnsure the kameleoonSimulationFFData cookie follows this format:
  • kameleoonSimulationFFData={"featureKey":{"expId":10,"varId":20}}: Simulates the variation with varId of experiment expId for the given featureKey.
  • kameleoonSimulationFFData={"featureKey":{"expId":0}}: Simulates the default variation (defined in the Then, for everyone else in Production, serve section) for the given featureKey.
⚠️ To ensure the cookie value works correctly, encode it as a URI component using a method such as encodeURIComponent.
CookieAccessor interface
Implement kameleoon::CookieAccessor to bridge SDK cookie reads and writes to your HTTP framework:
  • get() must return a cookie already written through set() on the response in preference to the request cookie of the same name. The SDK keeps no cookie state between calls, so this is what makes a second get_visitor_code() call within the same request return the code written by the first one instead of generating a new one.
  • The overrides are invoked from inside the SDK and must not throw. An exception escaping get() is treated as an absent cookie. Cookie values must be valid UTF-8.
Parameters
Return value
Exceptions

add_data()

The add_data() method adds targeting data to storage so other methods can use the data to decide whether or not to target the current visitor. The add_data() method doesn’t return any value and doesn’t interact with Kameleoon backend servers on its own. Instead, the SDK saves all the declared data for future transmission using the flush() method. This approach reduces the number of server calls made, as the data is typically grouped into a single server call that’s triggered by the flush(). The track_conversion() method also sends out any previously associated data, just like the flush(). The same holds true for get_variation() and get_variations() methods if an experimentation rule is triggered.
Each visitor can only have one instance of associated data for most data types. However, CustomData is an exception. Visitors can have one instance of associated CustomData per index.
The data parameter is a std::vector<Data>, where Data is a std::variant of all the data types. Each data type converts implicitly to Data, so you can pass a braced list of data objects.
Parameters
Exceptions

flush()

  • 📨 Sends tracking data to Kameleoon
The flush() method aggregates all Kameleoon data associated with a visitor and sends a tracking request to the server. This request includes any data previously added via the add_data method that hasn’t yet been transmitted through other tracking mechanisms (see the referenced methods for details). The flush() operation is non-blocking, as the server call is performed asynchronously. This method provides control over when data linked to a specific visitor_code is transmitted. For example, if add_data() is called multiple times, sending a request after each invocation would be inefficient. Instead, you can batch these updates and call flush() once to send all accumulated data in a single request. The flush() method uses the provided visitor_code as the unique visitor identifier.
  • flush(): Queues a flush operation according to the configured tracking interval.
  • flush_instant(): Sends tracking data immediately without waiting for the interval.
Parameters
Exceptions

get_remote_data()

The get_remote_data() method lets you retrieve remote data stored on Kameleoon servers for the specified key. In most setups, this data is written through the Kameleoon Data API and can later be fetched by your C++ service whenever you need additional app context. This method is useful when you want to keep structured information on Kameleoon’s remote infrastructure and reuse it from your back-end without maintaining a separate retrieval mechanism.
Parameters
Return value
Exceptions

get_remote_visitor_data()

get_remote_visitor_data() retrieves Kameleoon visit data for the provided visitor_code. The method adds the data to local visitor storage so other SDK methods can use it for targeting decisions. Data obtained using this method is especially useful when you want to:
  • use data collected from other devices.
  • access a visitor’s history, such as previously viewed pages from past visits.
  • use data that’s only available on the client side, such as datalayer variables and front-end goal conversions.
Read this article for a better understanding of possible use cases.
By default, get_remote_visitor_data() automatically retrieves the latest stored custom data with scope=Visitor and attaches it to the visitor without the need to call add_data(). This is especially useful for synchronizing custom data across multiple devices.
Parameters
Exceptions
Using parameters in get_remote_visitor_data()
The get_remote_visitor_data() method lets you control which data is retrieved for a visitor. The same filtering approach works across goals, experiments, variations, and other visitor data. For example, if you want to target users who converted on a goal during their last five visits, you can set previous_visit_amount to 5 and conversions to true. The flexibility shown in this example isn’t limited to goal data. You can use the filter to retrieve many different visitor behaviors and make them available to targeting and reporting logic in your C++ app.
RemoteVisitorDataFilter fields

get_visitor_warehouse_audience()

This method retrieves audience data associated with a visitor in your warehouse integration by using the specified visitor_code and, optionally, a warehouse_key. The warehouse_key is typically your internal user ID. The custom_data_index parameter corresponds to the Kameleoon custom data that Kameleoon uses to target your visitors. When the call succeeds, the SDK converts the returned audience list into CustomData, adds it to the visitor locally, and makes it available for targeting purposes. For more background, see the warehouse targeting documentation.
Parameters
Exceptions
You must use this method to specify whether the visitor has given legal consent to use personal data. Setting consent to false limits the types of data that can be included in tracking requests. This helps you adhere to legal and regulatory requirements while responsibly managing visitor data. For more information, see the consent management policy. If you pass a CookieAccessor, the SDK also updates the visitor cookies according to the consent status.
Parameters
Exceptions
Consent revocation behavior
When you call set_legal_consent() with consent=false, the SDK doesn’t delete the kameleoonVisitorCode cookie. Instead, it stops extending the cookie’s expiration date, allowing the cookie to persist until it naturally expires. If your compliance requirements demand the immediate removal of the cookie file upon opt-out, you must delete it manually using your framework’s native cookie management methods. The SDK won’t remove the file automatically.

Goals and third-party analytics

track_conversion()

  • 📨 Sends Tracking Data to Kameleoon
Use this method to track a conversion for a specific goal and user. This method requires visitor_code and goal_id. In addition, this method also accepts an optional revenue, negative and metadata arguments. The visitor_code is usually identical to the one you used when triggering the experiment.
This method is non-blocking because the SDK makes the server call asynchronously.
Parameters
The optional revenue, negative, and metadata parameters are passed as the fields of a TrackConversionOptions struct.
metadata values are accessible through raw data exports and the results page.If you provide the metadata parameter, Kameleoon uses these specified values for the current conversion instead of what you previously collected using the add_data() method. If you omit the parameter, Kameleoon uses the last tracked values for those CustomData prior to the conversion and within the same visit.Kameleoon will only consider the metadata values that are explicitly passed as parameters to the track_conversion() method.In the example below, Kameleoon will associate the conversion only with the custom data value explicitly provided as a parameter (here: index 5 with the value ‘Amex Credit Card’).
Exceptions

get_engine_tracking_code()

Kameleoon integrates with several analytics solutions, including Mixpanel, Google Analytics 4, and Segment. To track server-side experiments correctly, call the get_engine_tracking_code() method after the visitor triggers an experiment. The SDK returns JavaScript queue commands for the experiments that the visitor triggered during the previous 5 seconds. When you insert this code into the page, Engine.js processes the commands and sends the exposure events through the active analytics integration. Refer to hybrid experimentation for more information on implementing this method.
  • To use this feature, implement both the C++ SDK and Kameleoon Engine.js. Because this flow uses Engine.js only for tracking, you can install the asynchronous tag before the closing </body> tag.
  • If you only want to track experiments in Kameleoon and don’t need to send exposure events to third-party analytics tools, use the JavaScript / TypeScript SDK. This option works well for serverless edge compute platforms. The JavaScript / TypeScript SDK automatically tracks variations when you call getVisitorCode, as long as you add the corresponding experiment assignments to window.kameleoonQueue.
  • You can insert the returned tracking code directly into an HTML <script> tag.
In this example, 123456 and 234567 are experiment IDs, and 7890 and 8901 are variation IDs. In your implementation, the SDK generates these values in the returned tracking code.
Parameters
Return value
Exceptions

Events

set_event_handler()

Use this method to register a handler for SDK events. The SDK calls the handler when the selected event occurs. The client keeps at most one handler per event type: registering a new handler for the same event type replaces the previous handler. The EventHandler constructor you use (EventHandler::datafile_update() or EventHandler::http_request()) selects the event type, so a handler and the events it receives can’t get out of sync. Passing an empty std::function ({}) removes the current handler for the selected event type.
DataFileUpdateEvent contains information about an SDK data file update.
Handlers are called synchronously on the SDK’s single worker thread, so they must not block: time spent in a handler delays data file polling and tracking flushes. They must be thread-safe and must not throw: an exception escaping a handler is caught and discarded, and never affects SDK behavior. Don’t call blocking SDK operations from a handler; use the *_async forms and finish the work on an application thread instead.An invocation already in progress may finish after you clear the handler or destroy the client, and the handler (with its captures) is destroyed on an SDK thread once it’s no longer in use. Keep captured objects alive and thread-safe accordingly. A handler may clear or replace itself.All KameleoonClient instances created for the same site code and environment share one handler per event type: the last handler set wins, and destroying any of those instances clears it. Keep a single client instance per site code and environment.
Parameters

Data types

This section lists the C++ data types declared by the SDK in the kameleoon namespace (kameleoon/data/*.hpp). All of them are plain structs that you pass to add_data().

ApplicationVersion

ApplicationVersion represents the semantic version number of your application.
A visitor can have only one ApplicationVersion. Adding a second instance will overwrite the first one.

Browser

The Browser data set stored here can be used to filter experiment and personalization reports by any value associated with it.

Conversion

The Conversion data set stored here can be used to filter experiment and personalization reports by any goal associated with it.
  • Each visitor can have multiple Conversion objects.
  • You can find the goal_id in the Kameleoon app.
Cookie contains information about the cookies stored on the visitor’s device.
Each visitor can only have one Cookie. Adding a second Cookie overwrites the first one.

CustomData

CustomData enables the association of any type of data with each visitor, making it an effective tool for targeting conditions in segments. Additionally, it can be used as a filter or breakdown in experiment reports. For more information about custom data, refer to this article. Define custom data types in the Kameleoon app or the Data API and use them from the SDK. CustomData has two constructors: one identifies the custom data by its id (index), the other by its name.
  • Each visitor is allowed only one CustomData for each unique id(name). Adding another CustomData with the same id(name) will replace the existing one.
  • The custom data ‘index’ can be found in the Custom Data dashboard under the “INDEX” column.
  • To prevent the SDK from sending data with the selected index to Kameleoon servers for privacy reasons, enable the option: Use this data only locally for targeting purposes when creating custom data.
  • Adding a CustomData instance created with a name when the SDK instance isn’t initialized or the name isn’t registered, will result in the data being ignored.

Device

You can use device data to filter experiment and personalization reports by any associated value.

Geolocation

Geolocation contains the visitor’s geolocation details.
  • Each visitor can have only one Geolocation. Adding a second Geolocation overwrites the first one.

OperatingSystem

OperatingSystem contains information about the operating system on the visitor’s device.
Each visitor can only have one OperatingSystem. Adding a second OperatingSystem overwrites the first one.

PageView

Store page view events.
The referrer index is available in the Kameleoon app on the acquisition channel configuration page. Be careful: the index starts at 0, so the first acquisition channel you create has the ID 0, not 1.

UniqueIdentifier

If you don’t add UniqueIdentifier for a visitor, visitor_code is used as the unique visitor identifier, which is useful for cross-device experimentation. When you add UniqueIdentifier{true}, the SDK links flushed data with the visitor associated with the specified identifier. This can be useful in situations where you can’t access the anonymous visitor_code originally assigned to a visitor, but you do have access to an internal identifier connected to that visitor through session merging.

UserAgent

Server-side experiments are more likely to be affected by bot traffic than client-side experiments. Kameleoon uses the IAB/ABC International Spiders and Bots List to recognize known bots and spiders, and it also uses the UserAgent field to filter out other unwanted traffic that might distort your conversion metrics. For more information, see the help article on bot filtering. If you use internal bots, send the user-agent value curl/8.0 to exclude them from analytics.

Returned types

DataFile

The DataFile contains the SDK configuration details. It can be extended with additional information if required by clients. If you need more details, contact your Customer Success Manager.

FeatureFlag

The FeatureFlag represents a set of properties that define a feature flag itself (for example, its Variations, Rules, environment status, and other related details). It can be extended with additional information if required by clients. If you need more details, contact your Customer Success Manager.

Rule

The Rule represents a set of properties that define a rule itself (for example, its Variations). It can be extended with additional information if required by clients. If you need more details, contact your Customer Success Manager.

Variation

Variation contains information about the visitor’s assigned variation, or the default variation when no specific assignment exists.
  • Variation describes the assigned or default variation, while Variable contains the details of each individual variable.
  • id and experiment_id can be std::nullopt, which indicates a default variation that’s not tied to a specific experiment assignment.
Additional helper methods:

Variable

Variable contains information about a variable associated with the assigned variation.
JsonValue
JsonValue represents the value of a variation variable in C++. The kind field tells the string-carrying kinds apart; value is a std::variant<bool, double, std::string>. Typed accessors return std::nullopt when the value is of another kind: