Apex — Training Data Summary

Version 1.0 · Published 1 October 2026

This summary describes the content Callstack used to post-train Apex. It follows the structure of the European Commission's Template for the Public Summary of Training Content for general-purpose AI models (July 2025).

Scope. Apex is a modification of a general-purpose AI model released by Qwen (Alibaba Group). This summary covers only the content Callstack added in post-training. The content used to train the base model is described in the documentation published by Qwen.

Status. On the basis of the compute used for our modification, measured against the indicative criterion in the Commission's Guidelines on the scope of obligations for providers of general-purpose AI models (one third of the training compute of the original model), Callstack is not the provider of a general-purpose AI model in respect of Apex. We publish this summary voluntarily, because we believe customers and rightsholders should be able to see what went into the model they use.

1. General information

1.1 Provider

1.2 Model

1.3 Training content — size and characteristics

Characteristics. Source code in JavaScript, TypeScript, Kotlin, Java, Swift, Objective-C, Objective-C++ and C++, together with technical documentation, READMEs and code comments. Natural-language content is predominantly in English.

Latest date of data collection: September 2026.

2. List of data sources

2.1 Publicly available datasets

None. Callstack did not use any pre-compiled public dataset.

2.2 Private datasets licensed from third parties

None.

2.3 Content collected from online sources

Callstack obtained source code and documentation directly from public open-source repositories and package registries. No general web crawling was carried out.

Selection. The repositories were selected deliberately rather than crawled: the React and React Native core repositories, and the most widely used open-source libraries and packages in the React Native ecosystem, chosen by adoption and download volume.

Principal repositories:

• React — github.com/react/react — MIT License
• React Native — github.com/react/react-native — MIT License

Libraries and packages. Open-source React Native libraries and packages, each distributed under the MIT License. The complete list of repositories, with their licences, is available on request from apex@callstack.com.

Domains from which content was obtained:

Content obtained: source code files, documentation files, READMEs, licence files and in-repository examples. Commit histories, issue trackers, pull request discussions and other user-generated discussion content were not included.

2.4 User data

None. Callstack processes all customer API traffic under zero data retention: prompts and outputs are never stored, and so cannot be used for training.

2.5 Synthetic data

Instruction and task data derived from the sources in Section 2.3 and 2.6 was generated to teach the model to apply that material to development tasks. This synthetic data was generated using the base model and Callstack's own tooling, and was not produced using any third-party model whose terms prohibit its use for model training.

2.6 Other sources

Callstack internal documentation. Technical documents written by Callstack describing approaches to greenfield and brownfield React Native development and to migrations between frameworks, architectures and versions. Callstack owns this content. It was reviewed before use to exclude client-confidential information and personal data.

3. Relevant data processing aspects

3.1 Respect of reservations of rights from text and data mining

All third-party content used in post-training was obtained from repositories whose licences — in every case the MIT License — expressly permit use, copying, modification and the creation of derivative works. Callstack relied on those licences as the basis for use, and did not rely on the text and data mining exceptions in Articles 3 and 4 of Directive (EU) 2019/790.

Callstack did not include content from any repository whose licence does not permit such use, or which is published without a licence. No content was collected by crawling; accordingly, no robots.txt or other machine-readable opt-out mechanism was relevant to collection. Callstack maintains a policy for compliance with Union copyright law, available on request.

Rightsholders who believe their content has been used contrary to their rights may contact apex@callstack.com.

3.2 Removal of illegal content

The post-training content consisted of source code and technical documentation from established, maintained open-source projects and Callstack's own documents. The sources were selected individually, which by design excludes the categories of illegal content — including child sexual abuse material and terrorist content — that the removal measures in the Template are aimed at. Content was additionally screened for credentials, secrets and personal data before use.

3.3 Other information

Attribution and licence notices. The MIT License requires that its copyright and permission notice be included in copies or substantial portions of the licensed software. Apex is trained to generate new code rather than reproduce its training material, but it may occasionally output passages closely resembling code from the sources above. Users are responsible for reviewing Outputs and complying with any licence obligations that apply to material they incorporate.

Updates. This summary will be updated whenever Apex is further trained on content that changes the information above.