Today's trending projects share a common thread: open-source alternatives to proprietary AI infrastructure are graduating from proofs of concept into production-ready systems. Memory, voice, video generation, and cross-platform research are all getting serious tooling that keeps data local and doesn't force vendor lock-in.
lfnovo/open-notebook: A Self-Hostable NotebookLM
Open Notebook is a privacy-focused alternative to Google's NotebookLM that lets you organize research, generate AI-powered insights, and produce multi-speaker podcasts from your own documents — without any data leaving your servers. It supports 18+ AI providers including OpenAI, Anthropic, Ollama, and LM Studio, so teams aren't locked into a single vendor. With 26.8k total stars and 794 added today alone, it's the most momentum-heavy project on trending and clearly filling real demand for self-hosted research tooling.
github.com/lfnovo/open-notebook
MemPalace/mempalace: AI Memory That Stays on Your Machine
MemPalace stores conversation history as verbatim text and retrieves it via semantic search, achieving 96.6% recall on LongMemEval without making a single external API call. It organizes memories into a structured hierarchy — wings, rooms, and drawers — supports four pluggable vector backends including ChromaDB and pgvector, and ships 29 MCP tools for direct integration with Claude Code and compatible AI assistants. The local-first design is the headline: persistent AI memory with no third-party routing, backed by benchmarks rather than marketing claims.
github.com/MemPalace/mempalace
mvanhorn/last30days-skill: Research Any Topic Across the Entire Web
This agent skill queries Reddit, X, YouTube, TikTok, Instagram, Hacker News, Polymarket, GitHub, Bluesky, Threads, and general web search simultaneously — then merges duplicate stories, scores results by real engagement metrics, and produces a cited summary brief. The zero-config core works immediately for Reddit, HN, Polymarket, and GitHub; additional platforms unlock with optional API keys. At 29.4k stars, it reflects growing demand for agent skills that can give AI systems a wide-angle view of current events rather than narrow, curated feeds.
github.com/mvanhorn/last30days-skill
microsoft/VibeVoice: Hour-Long Audio in a Single Pass
Microsoft's open-source voice framework handles transcription and synthesis at a scale most models can't touch: 60 minutes of long-form audio in a single pass, preserving speaker identity and timestamps throughout. The TTS side generates up to 90 minutes of multi-speaker conversation between four distinct voices, while the streaming variant achieves roughly 300ms latency for real-time use. With 48.6k total stars, support for 50+ languages, and a lightweight 0.5B streaming model, VibeVoice is positioning itself as the foundation for the next generation of voice-enabled agents.
github.com/microsoft/VibeVoice
heygen-com/hyperframes: Write HTML, Render Video
HyperFrames lets developers produce deterministic MP4 videos using web technologies they already know — plain HTML, CSS, GSAP animations, Lottie, Three.js — with no video editing tools required. The framework is explicitly designed for agents: Claude, Cursor, and Gemini can author video compositions through bundled skills, and identical inputs always produce identical frames, making it CI/CD-friendly. Built by HeyGen, it targets automated product demos, PR walkthroughs, and documentation videos, and at 25.2k stars it's already become a go-to tool for teams that want video output without a video production team.