What's Trending on GitHub: March 2026

The March 2026 Leaderboard

Source: GitHub trending, Trendshift.io, ByteByteGo. Data as of March 29, 2026.

Pattern 1: Local-First Is Winning

OpenClaw (263K stars) runs your personal AI assistant entirely on your machine. Ollama (162K stars) lets you run Llama, Mistral, Gemma, and dozens of other models with a single command. GitNexus (35.9K stars) builds knowledge graphs client-side with no server dependency.

The thread connecting these repos: data sovereignty. Every one of them lets you use AI without sending a single byte to a third-party server.

This isn't just a developer preference — it's a market signal. Enterprise buyers increasingly require on-premise or local-first AI due to GDPR, HIPAA, SOC 2, and the EU AI Act (effective August 2026). The repos that are gaining stars fastest are the ones that make compliance the default, not an add-on.

Pattern 2: Open-Source AI Infrastructure Over Wrappers

Notice what's noton the list: thin API wrappers around GPT-4 or Claude. The days of “we call the OpenAI API and add a nice UI” repos going viral are over.

What's trending instead is infrastructure. Dify (130K stars) is a complete AI app builder platform — prompt engineering, RAG pipelines, agent orchestration, observability, all in one open-source package. n8n (150K+ stars) is workflow automation that connects AI to everything else in your stack.

The market is maturing. Developers don't want another chatbot interface. They want the plumbing — the systems that let them build, deploy, and monitor AI applications at production scale.

# Dify: one command to run a full AI app platform
docker compose up -d

# n8n: workflow automation with 400+ integrations
npx n8n start

# Ollama: run any model locally
ollama run llama3.2

All three commands run locally. All three are open source. All three reached 100K+ stars by solving infrastructure problems, not by wrapping someone else's API.

Pattern 3: The Agent Plugin Ecosystem Is Exploding

The Claude Skills collection hit 119K stars this month. Superpowers (by obra) reached 113.5K. The MCP server ecosystem now has 8,600+ servers — up 873% from mid-2025.

This is the “app store moment” for AI agents. Instead of building monolithic AI applications, developers are building small, composable skills that any agent can use. A SKILL.md file works across 11 different AI coding tools. An MCP server works across Claude, Cursor, Windsurf, and any client that speaks the protocol.

The implication for developers: the value is shifting from building standalone AI apps to building plugins that plug into the agent ecosystem. The repos that are growing fastest are the ones that make it easy to create, share, and discover these plugins.

What to Watch in April

How Much AI Code Is in These Repos?

We scanned several of these trending repos with GitIntel. The results vary wildly. Ollama has 0.8% AI-attributed commits. Dify is likely higher but uses tools that don't leave attribution markers.

The fastest-growing repos are also the ones most likely to include AI-generated code — and the least likely to track it. As these projects scale, understanding what percentage of their codebase is AI-generated becomes a supply chain question, not just a curiosity.

Want to check for yourself? Point GitIntel at any repo and see the data.