Projects Recently Starred by Experts

Starred 1 hour ago by

Jason Miller(Author of Preact). Also starred by

Eric Zhang(Founding Engineer at Modal),

Jonathan Ragan-Kelley(Professor at MIT),

Chris Van Pelt(Cofounder of Weights & Biases), and

3 more.

agentfs by tursodatabase

1.0%

Filesystem and storage abstractions for AI agents

Provides POSIX-like filesystem, key-value store, and tool call audit trails for agent state.
Enables auditability, reproducibility, and portability using a single SQLite database file.
Offers SDKs (TypeScript, Rust) and CLI for programmatic and command-line agent filesystem management.
Includes an experimental Linux sandbox for controlled agent execution and security enforcement.

Created 4 months ago

Updated 1 week ago

Starred 1 hour ago by

Abubakar Abid(Cofounder of Gradio).

llmfit by AlexsJones

128.0%

LLM model selection and optimization tool for local hardware

Right-sizes 94+ LLM models to your system's RAM, CPU, and GPU for optimal local inference.
Supports multimodal models, dynamic quantization, and MoE architectures with intelligent VRAM estimation.
Employs multi-dimensional scoring (quality, speed, fit, context) and backend detection (CUDA, Metal, ROCm) for tailored recommendations.
Integrates as an agentic framework skill for hardware-aware model selection and auto-configuration.

Created 2 weeks ago

Updated 14 hours ago

Starred 6 hours ago by

Eric Zhu(Coauthor of AutoGen; Research Scientist at Microsoft Research). Also starred by

Elie Bursztein(Cybersecurity Lead at Google DeepMind) and

Jared Palmer(SVP at GitHub; Founder of Turborepo; Author of Formik, TSDX).

OpenSandbox by alibaba

208.3%

Sandbox platform for AI and LLM applications

Provides multi-language SDKs and unified protocols for diverse AI scenarios.
Enables LLM capabilities: command execution, file operations, and agent execution.
Integrates Docker and Kubernetes runtimes for scalable, distributed sandbox scheduling.
Supports browser automation, VNC desktop, and VS Code web environments.

Created 2 months ago

Updated 12 hours ago

Starred 10 hours ago by

Travis Fischer(Founder of Agentic). Also starred by

Wing Lian(Founder of Axolotl AI),

Li Jiang(Coauthor of AutoGen; Engineer at Microsoft),

Nir Gazit(Cofounder of Traceloop), and

10 more.

superpowers by obra

13.8%

67k

AI assistant superpowers via a comprehensive skills library

Enhances AI agents with systematic debugging, TDD, and collaborative workflows.
Integrates slash commands for planning, execution, and interactive design refinement.
Facilitates parallel agent dispatching and semantic search of past conversations.
Promotes skill creation and contribution using TDD and evidence-based practices.

Created 4 months ago

Updated 1 week ago

Starred 13 hours ago by

Chaoyu Yang(Founder of Bento). Also starred by

Jeremy Howard(Cofounder of fast.ai),

Tim Davis(Cofounder of Modular), and

Jeff Hammerbacher(Cofounder of Cloudera).

llm.mojo by dorjeduck

0.3%

361

Mojo port of Karpathy's llm.c for GPT-2 training

Trains a GPT-2 model on the Tiny Shakespeare dataset.
Mojo implementation mirrors the original C version.
Includes a vectorized parameter update.
Includes tests for validation of the port's functionality.

Created 1 year ago

Updated 6 months ago

Starred 1 day ago by

Jeff Hammerbacher(Cofounder of Cloudera). Also starred by

Patrick von Platen(Author of Hugging Face Diffusers; Research Engineer at Mistral),

Yaowei Zheng(Author of LLaMA-Factory),

Pawel Garbacki(Cofounder of Fireworks AI), and

3 more.

ml-cross-entropy by apple

0.8%

584

PyTorch module for memory-efficient cross-entropy in LLMs

Reduces memory footprint by computing loss without materializing full logit matrix.
Uses custom Triton kernel with flash memory for matrix mult. and log-sum-exp.
Supports vocabulary parallelism for sharded classifier weights.
Includes optimized torch.compile implementation for CPU/older GPUs.

Created 1 year ago

Updated 5 months ago

Starred 1 day ago by

David Cournapeau(Author of scikit-learn).

Step-3.5-Flash by stepfun-ai

5.4%

Fast, efficient agentic intelligence model

Delivers frontier reasoning, agentic, and coding capabilities with high throughput for complex tasks.
Employs a sparse Mixture of Experts (MoE) architecture and Multi-Token Prediction for efficient inference.
Achieves 256K context window using Sliding Window Attention and supports local deployment via vLLM, SGLang, llama.cpp.
Achieves top performance in coding and reasoning benchmarks, rivaling proprietary models.

Created 1 month ago

Updated 3 days ago

Starred 1 day ago by

Chris Van Pelt(Cofounder of Weights & Biases). Also starred by

Theo Browne(Founder of Ping.gg),

Johannes Schickling(Cofounder of Prisma),

David Cournapeau(Author of scikit-learn), and

1 more.

cmux by manaflow-ai

71.7%

Parallel coding agent CLI manager

Run multiple AI coding assistants like Claude, Codex, and Gemini concurrently.
Spawns isolated VS Code workspaces for verifiable, fast agent task execution.
Supports cloud or local Docker environments for flexible deployment.
Integrates git diff, terminal, and dev server previews within agent workspaces.

Created 1 month ago

Updated 15 hours ago

Starred 1 day ago by

Caleb Peffer(Cofounder of Firecrawl). Also starred by

Travis Fischer(Founder of Agentic) and

Nicolas Camara(Cofounder of Firecrawl).

firecrawl-observer by firecrawl

0.8%

478

Website change monitoring with AI-powered analysis

Tracks website changes using Firecrawl API and Next.js
AI filters noise for intelligent change detection and notifications
Supports email, webhook, and dashboard-only monitoring options
Encrypts API keys with AES-256-GCM for enhanced security

Created 7 months ago

Updated 7 months ago

Starred 1 day ago by

Piero Molino(Cofounder of Predibase). Also starred by

Daniel Han(Cofounder of Unsloth) and

Michael Han(Cofounder of Unsloth).

mlx-tune by ARahim3

35.1%

617

LLM fine-tuning prototype for Mac users

Prototype LLM fine-tuning locally on Apple Silicon Macs using the MLX framework.
Seamlessly port training code from Mac prototypes to cloud GPUs with a Unsloth-compatible API.
Supports SFT, DPO, ORPO, GRPO, KTO, SimPO training methods via TRL and HuggingFace integration.
Leverages unified memory and exports models to HuggingFace or GGUF formats for broad compatibility.

Created 1 month ago

Updated 3 weeks ago

...