Skip to main content

AI features

Q247 has two features based on artificial intelligence. They answer different questions, are deployed independently, and you can use one without the other.

FeatureAnswers the questionNature
AI Adoptionhow much of the team's work is produced with the help of AI tools and how much it costsmeasurement, numeric data
AI Assistantwhat the data of this project showsinterpretation, text

AI Adoption as a measurement​

AI Adoption combines two sources: AI tool telemetry (work time, tokens, cost) and commit recognition based on the trailers such tools leave. This lets it answer the question of return on licenses, based on what was actually used, not on the number of purchased seats.

Commit recognition works on data Q247 collects anyway, so this part requires no additional configuration; the patterns themselves and recommendations regarding trailers are described in AI Adoption in Configuration. Telemetry requires configuring the export on the AI tool side and telling the plugin where to fetch it from, which is described in AI tool logs.

AI Assistant as interpretation​

The AI Assistant works on the project page. It retrieves project data from the last four weeks, passes it to the language model, and shows the result in descriptive form, together with conclusions and recommendations. The answer is cached for a few hours, so the next visit to the project shows it immediately.

The deployment has three variants, differing in who provides the language model: the Client's own model in the cloud, the Client's own model on-premise, or Q247 infrastructure. The choice affects network rules, the data processing agreement, and cost, so it is a deployment decision, not just a technical one. The details of each variant are described in Language model in Integrations.

Scope of data passed to the models​

Both features pass a different scope of data to the language models, which matters when the deployment is assessed by the security team.

AI Adoption passes nothing to language models. It processes telemetry and commit messages on the Q247 side.

AI Assistant passes to the model prompts and aggregated project data, including the first names, last names, and email addresses of team members. The source code of repositories is not passed. The full scope, together with the retention rules, is described in Security and data scope.

Subpages​

See also​