all-agentic-architectures vs awesome-llm-apps
35 agentic patterns (Reflexion, LATS, GraphRAG, MemGPT, Voyager…) as one Python library plus a runnable textbook — real LLM outputs, 9 providers, benchmark leaderboard, 283 tests. — versus — 141k-star collection of 100+ runnable AI agent, agent-skill and RAG app templates — clone one, pip install, run — across Claude, Gemini, GPT and open models. Apache-2.0.
Both teach agent patterns through runnable code; awesome-llm-apps is a gallery of standalone apps, all-agentic-architectures one tested library covering 35 named patterns.
| all-agentic-architectures | awesome-llm-apps | |
|---|---|---|
| Stars | 4.6k | 141k |
| Forks | 772 | 21k |
| Language | Jupyter Notebook | Python |
| License | MIT | Apache-2.0 |
| Last activity | 3 months ago | 11 days ago |
| Topics | agents | agents, rag |
| Curated connections | 2 | 2 |
all-agentic-architectures — the curator's take
The reference shelf for agent design, executable: every major pattern from the literature as an `Architecture` class with a uniform contract, each with a fully-run notebook (no mocked outputs) and a benchmark leaderboard ranking patterns per task — so 'which architecture for this job' gets an empirical answer, not a blog opinion. Read it before you invent your own loop. NOT a production framework: the uniform contract serves comparison and learning, and you'll re-implement the winner in your own stack — treat it as the textbook it says it is, not a dependency.
awesome-llm-apps — the curator's take
The fastest way to see a working agent or RAG pattern end to end and fork it: travel agents, research teams, voice agents, MCP apps, installable skills. Treat each folder as a starting point, not a dependency — most are single-file Streamlit demos with no shared library, no versioning and no production hardening. Want the patterns themselves as a reusable library with tests? all-agentic-architectures is the better shelf.