# Okaeri: Local Embedding and Ranking for Tech Feeds A privacy-focused feed aggregator utilizing local LLMs for personalized content discovery. - **Source:** [Hacker News: Show HN](https://news.ycombinator.com) - **Published:** 2026-10-20 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/5ac32c62-f7aa-4bf2-91bc-78cec24904b0 ## Personalized Feed Aggregation Okaeri serves as a minimal startpage and feed recommendation engine designed to replace cluttered interfaces like Skimfeed. The project originated from a personal Glance page that gained traction among colleagues. It focuses on providing a clean, user-controlled environment for consuming tech blogs and various feeds without the typical algorithmic noise found elsewhere. ## Local Embedding Architecture The system utilizes EmbeddingGemma 2, specifically the 270M 8-bit ONNX model, running directly on the CPU. By processing embeddings locally, the tool avoids external dependencies for content analysis. This architecture ensures that the recommendation logic remains entirely on the user's machine, prioritizing data privacy and reducing reliance on centralized cloud-based processing services. ## Custom Ranking Mechanism A custom ranking algorithm processes these embeddings alongside user interactions like clicks, likes, and hides. Preferences are segmented by topic, ensuring that high engagement with AI content does not suppress other categories. Users can further refine the output through manual slider adjustments, providing granular control over the relevance of displayed feed items. ## Resource Management Strategy To maintain system performance, the model operates within a short-lived child process. This approach keeps the server footprint light, with peak memory usage reaching approximately 1.3 GB during the embedding phase. Vectors are truncated to 256 dimensions and stored efficiently in SQLite, ensuring the application remains responsive even with continuous feed updates. ## Cross-Platform Compatibility Okaeri is released under the AGPL license and supports macOS, Linux, and Windows environments. While Windows support is a recent addition, the installation process is streamlined into a single command. This accessibility allows developers to quickly deploy the engine locally and begin customizing their feed consumption experience with minimal setup time required. ## Future Syncing Capabilities Development plans include creating a self-hosted version that syncs seamlessly with mobile devices. The creator is exploring methods to host a public version where embeddings are processed globally while maintaining local click data and ranking. This hybrid model aims to balance performance efficiency with the core requirement of keeping user preferences private. ## Social Media Integration A potential future feature involves syncing the engine with external social media feeds like Twitter or LinkedIn. By routing these platforms through the local recommendation engine, users could regain control over their content consumption. This mechanism is intended to mitigate doomscrolling by applying the same local ranking logic to broader social network streams. ## Open Source Collaboration Currently in its early stages, the project is open to community feedback and contributions. The developer explicitly avoids implementing account-based systems, seeking alternative ways to manage data synchronization. This transparent approach invites developers to inspect the code, suggest improvements, and help shape the future of local, user-centric feed recommendation technologies. ## Sources covering this story - https://github.com/DBvdHeijden/okaeri — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._