awesome-llm-apps: 100+ Runnable AI Agent & RAG Apps

Background and Context

Building a large language model application is rarely blocked by a single API call. The hard part is turning an idea into a complete, working program. Developers face a fragmented landscape. Some tutorials cover only prompts. Some frameworks offer only abstract interfaces. Few resources give an end-to-end example that a reader can run and then change. The awesome-llm-apps repository targets exactly this gap.

The project is maintained by Shubhamsaboo on GitHub. It has 142,327 stars. Its description is plain: more than 100 AI agents, Agent Skills and RAG apps, free and open source. It uses the Apache-2.0 license. The README says the templates are hand-built and tested end to end. It also invites readers to clone them, deploy them, and even sell them.

The collection does not tie itself to one model. Claude, Gemini, GPT, DeepSeek, Llama and Qwen are all in scope, both closed and open weights. This makes the repository a library of runnable cases, not another framework. A framework supplies the bricks. This collection shows what the finished house looks like.

Deep Analysis

The repository is split by purpose. The first block is Agent Skills. Their goal is to add new abilities to coding agents. One command installs a skill, and a reader calls it in everyday language. The skills work with Claude Code, Codex, Cursor and similar tools. The README states that each skill ships with real code and must pass security and evaluation CI checks. That gate is uncommon. Many collections list only prompt text and do no quality control. The skill topics are specific. Project Graveyard finds abandoned side projects and analyzes why each one died. First Reader simulates a real reader going through your draft. It reports where interest fades and where the reader stops, and it does not rewrite a word. Scope Creep Detector checks whether a diff goes beyond its stated intent. It then suggests whether to keep, split or explain the change. Commit Archaeologist rebuilds the reason a piece of code exists from its introducing commit, later edits and co-changes. Dependency Doctor inspects a dependency manifest for pinned standard-library versions, outdated backports, unpinned entries, duplicate constraints and yanked releases.

The second block holds standalone apps and tools. Excaliflow draws system architecture diagrams with animated arrows on an Excalidraw-style canvas and exports them as video. Needle is a Chrome extension. It searches the web by meaning and highlights the sentences from the most relevant sources. Other entries include a live insurance-claims agent team, an AI fraud-investigation agent, a home-renovation agent and an always-on HN briefing agent. These cover multi-agent teamwork and long-running operation. How do you run an awesome-llm-apps template? The path is short. To add a skill to a coding agent, run npx skills add and point it at the matching folder, such as project-graveyard. Then ask a plain question, for example why you never finish side projects. To run a full app, clone the repository and open the ai_travel_agent folder under starter_ai_agents. Install the packages in requirements.txt. Then start travel_agent.py with streamlit run. The README says this takes under 30 seconds.

Industry Impact

The value for the developer community is that "can it be done" becomes "follow this and do it". Multi-agent systems, RAG and skills are often still ideas in blog posts and talks. A cloneable, runnable, permissively licensed set of examples shortens the road from understanding to a working build. The design of running first and reading code second suits three groups. Learners want to see real Agent and RAG code structure. Engineers need fast prototypes and can pick a close template to adapt. Teams want to add new skills to their coding workflows.

The documentation supports this use. Each template has its own folder. The README front page offers a category index. It links to step-by-step tutorials on Unwind AI. It also says new templates arrive every week, with an email subscription for updates.

Separating Agent Skills into their own category is worth noting. It suggests that coding agents are becoming a new distribution channel. Abilities are packaged as small installable units. The CI gates for security and evaluation build a minimum level of trust for that channel. Community interest is clearly high at 142,327 stars. Stars do not prove that every template is still maintained, so readers should check each folder's dependencies and update history before they adopt it.

Outlook

Several risks deserve a clear statement. First, the 100-plus templates depend on different models and third-party services. Version drift will make older templates fail over time. Second, example code suits learning and prototyping. Before production use, a team must run its own security review, cost control and error handling. Third, skills touch your repository and commit history. Read a skill's code and permissions before you install it.

Three questions are worth watching. Will the evaluation standards for skills become more transparent? Can meta-loop skills that coordinate several models, such as Advisor Orchestrator Worker, settle into stable patterns? And can the weekly update pace keep quality steady while the count grows?

For now, the project gives developers a practical starting point. It is a runnable catalog of Agent, RAG and skill examples under the Apache-2.0 license. Its worth will depend on how well maintainers keep the templates current and how carefully users review what they install.