Awesome LLM Apps

What it is

Awesome LLM Apps is a GitHub repository of example AI apps and agent skills, under the Apache-2.0 license. Its README says they are hand-built and tested end to end. It is not one program. The README is an index, split into sections such as Agent Skills, Starter AI Agents, Advanced AI Agents, Voice AI Agents, MCP AI Agents and RAG. Most entries link to a folder in the repository holding one app or skill, and two link to other repositories and are marked external. The app folders these notes use each have their own README. An app uses the model provider its own README names, though the main README says the collection works with Claude, Gemini, GPT, DeepSeek, Llama, Qwen and other open-source models.

1. Pick a skill or an app in the README’s list
2. Open its folder and read its README
3. Install the skill, or clone and pip install the app
4. Enter the keys the app asks for
5. Run it and try the README’s example prompts

Who it suits

Developers who learn by running a working example and then changing it. People who use a coding agent and want to add a skill to it. Anyone who wants to see how a RAG pipeline, an MCP client or a team of agents is put together in Python.

Skip it if you want production software with support. The README calls its entries templates, and the Apache-2.0 license provides the software on an “AS IS” basis, without warranties or conditions of any kind.

Where to use it

The skills work in the coding agents the README names. Most apps are Python programs you start from a terminal. Some, such as the Generative UI starter, are npm projects with a Next.js frontend.

• Coding agents: Claude Code, Codex, Cursor, GitHub Copilot and VS Code, Antigravity, OpenClaw and Hermes, for the skills
• Your terminal and browser: the Streamlit apps start with streamlit run and open as a web page
• Local models: the Local RAG Agent runs Llama 3.2 through Ollama, with Qdrant in Docker, and its AgentOS page is typically at localhost:7777
• Docker: the GitHub MCP Agent runs the official GitHub MCP server in Docker
• Google ADK: the crash course starts its web interface with adk web

The maker’s coding-agent install steps cover the skills. Two app folders, the Generative UI starter and the dashboard canvas agent, also carry a CLAUDE.md or AGENTS.md for a coding agent working in them. For the Python apps, Mindket’s notes are plain steps you run in a terminal.

How to set it up

Get it saves a zip of Mindket’s setup notes in your downloads folder. The apps and skills install from the repository, not from the zip.

1. Pick a skill or an app in the README.
2. For a skill, run the first command below with the skill’s folder name.
3. For an app, clone the repository, change into the app’s folder and install its requirements.
4. Start it with the command in the app’s README, and enter the keys it asks for.

npx skills add https://github.com/Shubhamsaboo/awesome-llm-apps/tree/main/agent_skills/project-graveyard

The README’s quick start gives that skills command, with Project Graveyard as its example. It gives the next four lines for the AI Travel Agent. Swap in the folder and the last command of the app you pick.

git clone https://github.com/Shubhamsaboo/awesome-llm-apps.git
cd awesome-llm-apps/starter_ai_agents/ai_travel_agent
pip install -r requirements.txt
streamlit run travel_agent.py

The travel agent shows two password fields on its page, one for an OpenAI key and one for a SerpAPI key.

How to use it

With a skill, ask your agent in plain words. The README’s example for Project Graveyard is the question below.

why do I never finish my side projects?

With an app, enter the keys it asks for on its page. The data analysis agent’s README says to put the OpenAI key in the Streamlit sidebar. That agent takes a CSV or Excel file and answers questions about it. The GitHub MCP agent answers questions about a repository you name.

Example

You have a folder of old side projects and want to know which one to finish. You install the Project Graveyard skill, then ask your agent:

run the graveyard on ~/dev and ~/projects

Its README says the scan reads git metadata and not your code, runs on your machine and makes no network calls. Per its SKILL.md, the agent turns the output into a report. The report has four parts: a census, tombstones with each project’s cause of death, your patterns, and one project to bring back.

The limits are in the README. It was tested on macOS and Linux, and it does not mention Windows. A project with no git history is counted but not diagnosed.

For newcomers and experts

If you are new. An LLM app sends your question to a language model and shows the answer. An agent can also call tools, such as a web search. RAG means the app first looks up passages in your own files and hands them to the model, so the answer comes from your documents. A skill is a folder with a SKILL.md file that a coding agent reads when the task needs it. An API key is a secret string from a model provider, and providers may charge for use, so check the provider’s pricing page first. Start with one Starter AI Agent and ask it a single question before you read the code.

For experts. Each app is its own project. The travel, data analysis, Local RAG and GitHub MCP apps are built with the Agno framework, and their requirements list agno>=2.2.10. The Generative UI starter pairs a Next.js frontend with a LangGraph agent, as its CLAUDE.md describes. The crash courses cover Google ADK and the OpenAI Agents SDK.

The file agent_skills/registry.json lists seven skills, each with its description, install command and eval folder. Six of them also name a license and author. A GitHub workflow, skill-evals.yml, runs the skills’ evals, a lint, a security scan, trigger checks and a registry check on pull requests, and on pushes to main, that touch agent_skills/ or README.md.

Good to know

Most of the questions here are about keys, your data and the licenses of single folders.

• License: Apache-2.0 for the repository. Some folders under generative_ui_agents carry an MIT license of their own. The starter project, the dashboard canvas agent, the financial coach and the MCP app builder name Atai Barkai, and the shadcn component generator names Tyler Slaton. Read the license in the folder you copy from
• Keys and cost: an app calls a model provider with your key, and the provider may bill you. Type a key into the app, and never into a chat
• Your data: an app sends what you type, and the files you give it, to the provider you choose. The Local RAG Agent keeps its model on your machine through Ollama. The local travel agent runs its model there too, but it still searches through SerpAPI
• Skills: they run with your agent’s permissions. The README asks you to read each SKILL.md and every script before you install it
• Sensitive topics: the list includes apps for medical images, legal work, money and mental wellbeing. Mindket’s notes do not cover them, so read the app’s own README before you trust an answer
• External links: two entries point to other repositories and are marked external. Their licenses are their own
• Change: new templates are added over time, so a name or folder here may have moved
• Outbound: reading on GitHub loads GitHub. The npx command downloads the skills package from the npm registry the first time, and the address you give it points at GitHub
• Removal: delete the clone folder, or the skill’s folder in your agent’s skills directory. The README gives no uninstall steps, and pip packages stay until you remove them with pip