walid@portfolio:~/lab/open-source-agent-repos$
cd../lab
06ideaAug 2026

Two repos worth cloning before building an agent

Around 150 open-source agents and RAG apps — adapt one instead of starting from a blank project

Two GitHub repositories holding working agent projects across voice, customer support, research, RAG, coding and workflow automation. The point is not to read them — it is to find one close to what you need, open it in Claude Code, and describe your use case until it becomes yours.

AI agentsOpen sourceRAGMCPVoice AIClaude Codeawesome-llm-appsai-engineering-hub
What they are

Working projects, not a reading list

Both are collections of complete, runnable agent projects rather than tutorials about agents. Between them their READMEs link roughly 150 projects, and awesome-llm-apps describes itself as “100+ open-source AI agents, agent skills, and RAG apps”.

They split differently, which is why both are worth having: one organises by what the agent is, the other by how hard the project is to build.

The two

How each one is organised

RepositoryOrganised byWhat it covers
awesome-llm-appsagent typeAgent skills, starter and advanced agents, always-on agents, multi-agent teams, voice agents, generative UI, game-playing agents, MCP agents, RAG, agents with memory, fine-tuning, framework crash courses
ai-engineering-hubdifficultyBeginner through advanced: OCR and vision, chat interfaces, basic then advanced RAG, multimodal, agents and workflows, voice and audio, MCP and infrastructure, model evaluation, fine-tuning, production systems
Inside

The kinds of agent you will find

voice AIcustomer supportemail automationresearch and web browsinglead qualificationdocumentation and knowledge managementcoding and developmentdata analysisworkflow automation
How to use them

Adapt one rather than start blank

1

Find the closest match

Not the perfect one. Close enough that the shape of the problem is already solved.

2

Clone or download it

You want the whole project on disk, not a snippet copied out of a README.

3

Open it in Claude Code

From the project root, so it can read the structure rather than guess at it.

4

Explain your actual use case

Your business, your workflow, the thing it has to do differently. Specifics beat instructions here.

5

Ask it to adapt the agent

Editing something that already runs is a different job from writing one from nothing, and a much shorter one.

i
Why this is worth the detour

Starting from a project that already runs skips the part where you find out your architecture was wrong. The wiring, the framework choices and the failure handling are already made — you are changing behaviour, not inventing structure.

← previous
Codex CLI as a content pipeline, not a coding tool
next →
An option-trading advisor built from Claude skills
← all experiments