supabase/supabase ⭐ 113.6k
open-source Postgres backend with auth and realtime
• Provides a hosted Postgres database for web, mobile, and AI apps.
▶️ https://t.co/Qn9RSa0LdP
#supabase#postgres#opensource
vllm-project/vllm ⭐ 95k
fast, memory-efficient LLM inference and serving at scale
• Manages GPU memory efficiently using PagedAttention for attention key and value storage.
▶️ https://t.co/7klDY4IVaU
#vllm#llm#opensource#neuraldashnexusdotnet
wshobson/agents ⭐ 41.4k
unified plugins for seven AI coding assistants
• Supports seven different AI coding tools with unified plugins
▶️ https://t.co/9XkccShIhD
#agents#codingagent#opensource
Comfy-Org/ComfyUI ⭐ 140.1k
visual node graph for local AI generation
• Build image, video, audio, 3D, and text workflows using a visual node graph without writing code.
▶️ https://t.co/cmwiitNDEV
#ComfyUI#StableDiffusion#OpenSource
stablyai/orca ⭐ 95.7k
run a fleet of AI coding agents in parallel
• Run multiple agents like Claude Code, Codex, and Grok side-by-side in separate git worktrees.
▶️ https://t.co/7pW5iE3SZ6
#orca#codingagent#opensource
xerj-org/xerj ⭐ 7.2k
index your codebase so AI agents stop wasting tokens
• Automatically indexes code, docs, logs, PDFs, SQLite, and CSVs into typed searchable indices with no schema required
▶️ https://t.co/3gAMRJUvAj
#xerj#codingagent#OpenSource
lightseekorg/tokenspeed ⭐ 2.6k
production LLM inference built for agentic workloads
• Separates control plane (C++) from execution plane (Python) for correctness and fast iteration.
▶️ https://t.co/GEIiFP5pNu
#tokenspeed#llm#opensource
pytorch/pytorch ⭐ 104.6k
GPU-accelerated tensors and deep learning in Python
• Runs tensor computations on GPU or CPU, similar to NumPy arrays
▶️ https://t.co/hzcxl9WC71
#pytorch#deeplearning#opensource
Weekend, chill time!
Working on,
Nexus UI — everything you build, one canvas.
Terminals were built for tabs. This one is a Windows workspace where agents, shells, SSH, logs, and a browser sit on the same surface and stay where you left them.
Listening to Boom bap music on YouTube Premium, Cursor open with Five subagents walking the PRD.
Model of choice for the build today: Grok 4.7 Extra High Fast.
Thanks,
@cursor_ai@elonmusk@grok