A hands-on series exploring DeepAgents — a framework built on LangChain and LangGraph for building agentic AI systems that can plan, use tools, manage memory, and work autonomously.
This repo accompanies a learning series I'm posting on LinkedIn and the Hexaware Station H portal — one concept at a time, with runnable code for each.
Most agent tutorials stop at "here's a chatbot with a tool." This series goes further: filesystem access with permission rules, memory that persists across runs, skills that load on demand, subagents, human-in-the-loop approval, context management for long-running agents, and connecting to external tools via MCP.
Each lesson is a small, self-contained, runnable script — no notebook required.
| # | Topic | What it demonstrates |
|---|---|---|
| 01 | Basics | Minimum viable agent: one tool, one model, one question |
| 02 | Multiple Tools | How an agent chooses the right tool among several |
| 03 | Virtual Filesystem | Built-in file tools (ls, read, write, edit, glob, grep) |
| 04 | Filesystem Permissions | Declarative allow/deny rules on file paths |
| 05 | Streaming | Getting incremental updates instead of waiting for the final answer |
| 06 | Memory | Persistent context via AGENTS.md files |
| 07 | Skills | On-demand domain knowledge via SKILL.md (progressive disclosure) |
| 08 | Subagents | Spawning isolated child agents for parallel/heavy subtasks |
| 09 | Task Planning | Built-in todo tracking for long multi-step runs |
| 10 | Human-in-the-Loop | Pausing for approval before risky tool calls |
| 11 | Context Engineering | Managing token limits automatically in long-running agents |
| 12 | MCP Tools | Connecting external Model Context Protocol servers as tools |
- DeepAgents — the agent framework
- LangChain / LangGraph — underlying orchestration
- Groq — fast, free-tier LLM inference for learning without cost
- Python 3.11+
git clone https://github.com/Shorya22/DeepAgents.git
cd DeepAgents
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txtAdd a .env file with your Groq API key:
GROQ_API_KEY=your_key_here
Then run any lesson directly:
python 01_deepagents_basics.pyNew lessons get added as I work through them, each paired with a write-up on LinkedIn and Station H. Feedback, questions, and PRs are welcome.
MIT