This daily update tracks GitHub repositories gaining significant momentum, ranking them by percentage star growth over the last seven days to surface emerging technologies before they reach mass adoption. This methodology prioritizes rapid growth over absolute scale, highlighting Andee’s 834% surge in on-device privacy-focused agents and B Desktop Companion’s 778% rise in proactive desktop overlays. The current landscape shows a concentrated push toward specialized infrastructure, evidenced by LayerFS’s 343% gain in versioned filesystems and OpenFilm’s 375% growth in AI-driven video production tools.
| # | Repository | Stars Gained | Gain % | Overview |
|---|---|---|---|---|
| 1 | kuafuai/andee | 2302 | 834.1% | An on-device AI agent for Android and macOS that operates device interfaces, manages local memory, and coordinates tasks across multiple devices without relying on vendor clouds. |
| 2 | AhmadHassan-BTed/B | 2827 | 778.8% | An agentic desktop companion that uses semantic screen awareness and LLM inference to proactively assist users during their workflow. |
| 3 | openfilm/openfilm | 1837 | 375.7% | An open-source AI video agent and editor that enables coding agents to generate, edit, and render high-quality films using web technologies. |
| 4 | Ephemeral-AI-Lab/layerfs | 1544 | 343.9% | LayerFS provides an ephemeral, versioned filesystem substrate that allows AI agents to perform isolated, writable computations while maintaining a shared, durable history of all changes. |
| 5 | FengZhenfei/carrel | 342 | 282.6% | A self-hosted knowledge base for AI agents that automates document parsing, indexing, and retrieval using hybrid search and knowledge graphs. |
Andee (kuafuai/andee )
Andee is an autonomous AI agent designed to act as a resident intelligence within Android and macOS devices. Unlike traditional assistants, it operates locally using OpenAI-compatible models to interact with apps via accessibility services, manage a private local notebook, and execute complex multi-step scenes like subscription management or meeting transcription. It features cross-device synchronization through a peer-to-peer “friend” system and can even interface with physical hardware like robot dogs. This tool is intended for privacy-conscious users seeking an agent that maintains full control over its own memory and execution boundaries.
B Desktop Companion (AhmadHassan-BTed/B )
B is an agentic desktop companion designed to live as a persistent, proactive overlay on a user’s workstation. It utilizes semantic screen sensing and OCR to understand active window content, allowing it to offer contextual insights or documentation without explicit user commands. The system features a dedicated ‘Work Mode’ for non-intrusive monitoring and supports both cloud-based and local LLM inference. It is intended for power users seeking an autonomous digital assistant that integrates seamlessly into their existing desktop environment.
OpenFilm (openfilm/openfilm )
OpenFilm is an open-source framework designed to bridge the gap between AI coding agents and video production. It allows users to instruct agents like Claude Code or Codex to generate films as HTML-based web pages, which can then be refined using Studio, a local browser-based editor. Key features include timeline-based editing, support for CSS and WebGL animations, and seamless integration with third-party providers like ElevenLabs and OpenAI for media generation. It is particularly useful for developers and creators looking to automate video production workflows through natural language commands.
LayerFS (Ephemeral-AI-Lab/layerfs )
LayerFS is a specialized filesystem designed to serve as the computational substrate for multi-agent systems. It enables agents to mount ephemeral workspaces, execute commands, and automatically commit changes to a shared, decentralized store. Key features include recursive branching, ultra-optimized storage through content-addressing, and built-in conflict resolution for collaborative agent workflows. This makes it a practical tool for developers building autonomous agentic workflows that require instant bootstrapping and a complete, auditable history of every tool call and state change.
Carrel (FengZhenfei/carrel )
Carrel is a self-hosted infrastructure designed to provide AI agents with structured access to complex document collections. It automates the ingestion of diverse formats—including PDFs, spreadsheets, and source code—and utilizes hybrid retrieval combining vector, keyword, and graph-based search. By extracting entities and relationships into knowledge graphs, it allows agents to follow logical connections and cite specific source passages or images. This makes it a practical tool for developers building agents that must navigate large volumes of research, contracts, or technical documentation.