Dear freedesktop.org Community, I am proposing a Unified AI API, a native, system-level interface for AI tasks (e.g., text summarization, smart notifications) integrated into Linux desktop environments via `libai`, akin to `libnotify`. The API enables no-import-required AI access, with a system app (e.g., GNOME/KDE Settings) for users to select models (e.g., Llama, Mistral) and configure privacy. Key Features: - Global `AI` API with methods like `AI.generateText()`, `AI.generateNotificationContent()`. - Integration with `libnotify` for AI-driven notifications. - Support for open-source models via TensorFlow/PyTorch, with ONNX compatibility. - User-controlled settings for on-device vs. cloud processing. Relevance to freedesktop.org: This aligns with freedesktop.org?s mission to standardize desktop specifications, enhancing Linux?s AI capabilities and developer experience. It complements existing standards like `libnotify` and supports GNOME/KDE ecosystems. Details and Contribution: The proposal, API spec, and Linux-specific examples are available at: https://github.com/beamitpal/unified-ai-api We welcome feedback via GitHub Issues or this mailing list. Could we propose this for discussion at a freedesktop.org meeting or FOSDEM? Please advise on the next steps for standardization. Sincerely, Amit me at beamitpal.com Unified AI API Proposer -------------- next part -------------- An HTML attachment was scrubbed... URL: <https://lists.freedesktop.org/archives/freedesktop/attachments/20250802/5ae44102/attachment.htm>