📈 AI Open Source Trends 2026-04-04

[github-actions]

AI Open Source Trends 2026-04-04

Sources: GitHub Trending + GitHub Search API | Generated: 2026-04-04 00:10 UTC

---

AI Open Source Trends Report — April 4, 2026

Step 1: Filter Applied

Excluded from trending: openscreen (demo creation tool, non-AI core), fff.nvim (general file search), sherlock (OSINT tool). Retained 5 AI-relevant trending repos + 80 topic-search results.

---

1. Today's Highlights

Oh My codeX explodes with 3,047 stars—developers are racing to extend GitHub's Codex CLI with hooks, agent teams, and HUDs, signaling massive appetite for "agent harness" customization. Google's TimesFM time-series foundation model gains 916 stars, showing enterprise demand for specialized non-NLP foundation models. Onyx (1,852 stars) and prompts.chat (375 stars) highlight dual momentum: enterprise-ready AI platforms and grassroots prompt engineering communities. The Rust-powered fff.nvim—though general-purpose—reflects infrastructure performance obsession for AI agent tooling. Together, these reveal a market bifurcated between polished enterprise suites and hacker-friendly agent extensibility layers.

---

2. Top Projects by Category

🔧 AI Infrastructure

Project	Stars	Why It Matters Today
ollama/ollama	167,021	Local LLM runtime now supporting Kimi-K2.5, GLM-5, MiniMax—expanding beyond Western models for global deployment
vllm-project/vllm	75,180	High-throughput inference engine; critical infrastructure as model serving costs dominate AI budgets
dmtrKovalenko/fff.nvim	0 (+750 today)	Rust-powered file search optimized for AI agents—sub-millisecond latency for agent context retrieval
langchain-ai/langchain	132,260	"Agent engineering platform" rebrand signals pivot from chains to autonomous agent orchestration
0xPlaygrounds/rig	6,772	Modular Rust LLM framework—Rust's memory safety meets AI infrastructure demands

🤖 AI Agents / Workflows

Project	Stars	Why It Matters Today
Yeachan-Heo/oh-my-codex	0 (+3,047 today)	Top gainer: Codex CLI supercharger with hooks, teams, HUDs—"Oh My Zsh for AI coding"
Significant-Gravitas/AutoGPT	183,120	Reborn as accessible AI platform; 183K stars show sustained autonomous agent interest
langgenius/dify	135,664	"Production-ready agentic workflow"—enterprise visual builder competing with proprietary platforms
browser-use/browser-use	85,893	Web automation for agents; critical primitive as agents need real-world tool access
shareAI-lab/learn-claude-code	47,864	Educational "nano claude code" implementation—democratizing agent harness internals
activepieces/activepieces	21,543	~400 MCP servers for AI agents—MCP (Model Context Protocol) ecosystem acceleration
trycua/cua	13,374	Computer-Use Agent infrastructure: sandboxes, SDKs, benchmarks for desktop-controlling agents

📦 AI Applications

Project	Stars	Why It Matters Today
onyx-dot-app/onyx	0 (+1,852 today)	Enterprise AI platform—"works with every LLM" positioning captures multi-model enterprise strategy
open-webui/open-webui	129,906	User-friendly AI interface; 129K stars prove UX layer critical for LLM adoption
CherryHQ/cherry-studio	42,865	300+ assistants, unified frontier LLM access—personal AI productivity consolidation
zhayujie/chatgpt-on-wechat	42,739	CowAgent: "lighter than OpenClaw"—Chinese market's preferred agent deployment
saturndec/waoowaoo	10,795	Industrial AI film/video production—Hollywood-grade agent workflows for creative vertical

🧠 LLMs / Training

Project	Stars	Why It Matters Today
google-research/timesfm	0 (+916 today)	Google's time-series foundation model—expanding foundation model scope beyond language
huggingface/transformers	158,762	Definitive model framework; multimodal expansion (vision, audio) keeps it central
rasbt/LLMs-from-scratch	89,916	Educational PyTorch implementation—educational depth as competitive moat
jingyaogong/minimind	45,531	2-hour 64M-parameter GPT training—democratizing LLM training for experimentation
hiyouga/LlamaFactory	69,468	100+ LLMs/VLMs unified fine-tuning—efficiency essential as model diversity explodes

🔍 RAG / Knowledge

Project	Stars	Why It Matters Today
infiniflow/ragflow	77,053	"RAG + Agent capabilities"—knowledge retrieval evolving into active agent context
mem0ai/mem0	51,888	Universal memory layer—solving agent state persistence across sessions
run-llama/llama_index	48,273	"Document agent and OCR platform"—RAG vendor pivot to agent-centric positioning
milvus-io/milvus	43,590	Cloud-native vector DB; scalability as enterprise RAG default
thedotmack/claude-mem	44,946	Claude Code memory plugin—agent session continuity via AI-compressed context
VectifyAI/PageIndex	23,893	"Vectorless, reasoning-based RAG"—challenging embedding-heavy paradigms

---

3. Trend Signal Analysis

Explosive Category: "Agent Harness" Extensibility

Today's 3,047-star surge for oh-my-codex reveals a critical inflection: developers don't want black-box AI coding tools—they want composable, hackable agent infrastructure. This mirrors the 2010s DevOps transformation where "infrastructure as code" replaced proprietary tooling. The pattern repeats across learn-claude-code (47,864 stars, educational agent harness), everything-claude-code (136,058 stars, performance optimization system), and claude-mem (44,946 stars, memory persistence).

New Tech Stack Emergence: MCP (Model Context Protocol)

activepieces/activepieces explicitly markets "~400 MCP servers for AI agents"—MCP, Anthropic's open protocol for model-tool integration, is becoming the "USB-C for AI agents." This standardization wave enables the jackwener/opencli vision: "Make Any Website & Tool Your CLI" via unified AGENT.md integration.

Industry Event Connection: GitHub Codex CLI Launch Fallout

The timing aligns with GitHub's recent Codex CLI release. Rather than accepting GitHub's defaults, the community immediately built extension ecosystems—oh-my-codex for customization, ralph-claude-code for autonomous loops, AionUi as unified Cowork app. This mirrors VS Code's extension economy genesis.

Vertical Specialization Signal

google-research/timesfm (time-series) and waoowaoo (film production) demonstrate foundation models and agent workflows penetrating domain-specific verticals beyond general chat—enterprise and creative industries respectively.

---

4. Community Hot Spots

• oh-my-codex — Highest velocity today (3,047 stars). The "Oh My Zsh for Codex" positioning suggests a new category: shell ecosystem for AI agents. Watch for plugin marketplace emergence.

• onyx — Enterprise "works with every LLM" strategy captures multi-model procurement reality. 1,852 stars indicate strong product-market fit for vendor-agnostic AI platforms.

• timesfm — Google's time-series foundation model signals enterprise AI moving beyond chatbots to operational forecasting. Critical for supply chain, finance, IoT applications.

• PageIndex — "Vectorless, reasoning-based RAG" challenges the embedding-heavy status quo. If proven at scale, could disrupt $B vector database market with pure reasoning approaches.

• cua + OpenSandbox — Computer-Use Agent infrastructure race intensifies. Sandboxed desktop environments for agents becoming critical primitive; Alibaba's entry signals Chinese market acceleration.

---

Report generated from GitHub trending data, April 4, 2026. Star counts reflect snapshot timing; velocity indicators prioritize today's new stars where available.

---

This digest is auto-generated by agents-radar.