// the find
Shubhamsaboo/awesome-llm-apps
100+ AI Agent & RAG apps you can actually run — clone, customize, ship.
A monorepo of 100+ standalone LLM app templates covering agents, RAG pipelines, voice agents, MCP integrations, and fine-tuning recipes — each self-contained with its own requirements.txt and runnable in a few commands. It's aimed at developers who want a working starting point rather than reading documentation and assembling from scratch. Provider-agnostic: most templates swap between OpenAI, Claude, Gemini, and local models via a config change.
Each template is genuinely self-contained — no shared internal library to debug, just clone the subdirectory and run it. The breadth is real: you get everything from a basic RAG chain to a multi-agent travel planner with a Next.js frontend, Prisma, and a FastAPI backend. The MCP agents section is more current than most similar repos, with browser, GitHub, and Notion integrations that actually reflect the 2025 MCP ecosystem. Apache-2.0 with no telemetry or signup wall means you can fork it commercially without reading fine print.
The quality variance across 100+ templates is high — the more complex ones (travel planner with backend+client) are legitimately well-structured, but many starter agents are single-file Streamlit scripts with hardcoded prompts and no error handling, which is fine for learning but gives you nothing reusable for production. No shared abstractions across templates means if you want to combine ideas from two of them, you're copy-pasting and reconciling different dependency versions manually. The 'runs in 3 commands' claim breaks down for anything beyond the simple Streamlit apps — the multi-agent examples often need multiple services, env vars, and in some cases a database. Test coverage is essentially zero across the repo.