11 AI Agents in Python with LangGraph: Free Source Code and Course

Most AI agent tutorials stop at a chatbot that calls one tool. Real business agents need more: they must recover from errors, ask a human before risky actions, remember customers, process many documents in parallel and refuse unsafe requests. Learning all of that from scattered blog posts is slow.

That is why I put 11 complete agents in one place. Each one solves a real business problem and shows one important LangGraph pattern, and they are ordered from easy to advanced. You can read the spec, take the course, run the code and then change it for your own project or client.

Short · 1:36

• Runs on a laptop: databases, order systems, metrics and paging tools are replaced with local sample data
• Same structure everywhere: each agent.py has State, Tools, Nodes, Graph and CLI sections
• Free check before you spend: python agent.py --graph prints the agent’s diagram without calling the model
• Easy to customise: swap the functions in the Tools section for your real database or API

New to agents? Start with my explainer below, then read the official LangGraph documentation alongside the course.

Tutorial · 16:40

↑ Back to topNext: The 11 AI agents at a glance →

What each agent does

01. SQL Data Analyst

Turns a plain-English question into a read-only SQL query, runs it on SQLite, repairs the query if the database returns an error (up to 3 attempts) and explains the result.

02. CSV/Excel Data Analyst

Answers questions about a CSV, Excel or Parquet file. The model writes one pandas expression, a sandboxed tool runs it, and you get a validated result with a summary, table and caveats.

03. BI Dashboard Insights

Ranks the biggest metric changes in code, tests which dimensions explain them, computes confidence and writes a short executive summary.

04. Customer Support

Classifies each message (order status, returns, tech help, billing), routes it to the right path and hands over to a human with a summary when needed.

05. HR Resume Screener

Builds a scoring rubric from a job description, removes personal data before the model sees a resume, scores all candidates in parallel and writes a ranked report.

06. Finance Expense Auditor

Checks expense reports against policy rules, finds duplicates, and pauses for a finance manager's approval; the pause is saved so you can resume later.

07. Marketing Content

A writer drafts copy from a brief, checkers flag unsupported claims and banned words, an editor scores it, and the best-scoring draft wins.

08. Legal Document Reviewer

Reviews a long contract against a checklist section by section, in parallel, and verifies every quoted citation against the original text.

09. Healthcare Intake

Collects pre-visit information while removing personal data, screening for emergencies and blocking any diagnosis or treatment advice.

10. DevOps Incident Triage

Gathers metrics, logs and recent changes in parallel, ranks causes, computes severity in code and proposes a runbook action that a human must approve.

11. E-commerce Recommender

A shopping assistant that remembers preferences across visits, enforces budget and size rules in code and lets shoppers see or delete what it remembers.

In this live session I walk through all eleven agents and their code:

Live session · 33:37

↑ Back to topNext: Run your first agent →

1. Install Python 3.11+ and a code editor; my VS Code guide shows the setup.
2. Get an API key from the Anthropic Console.
3. Clone the repository (or download the ZIP from GitHub).
4. Set up one agent with the commands below, starting with 01_sql_data_analyst.

git clone https://github.com/AammarTufail/eleven_ai_agents_for_begineers_to_advance_level_developers.git
cd eleven_ai_agents_for_begineers_to_advance_level_developers/agents_code/01_sql_data_analyst
python3 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
pip install -r requirements.txt
cp .env.example .env        # then open .env and paste your ANTHROPIC_API_KEY

python agent.py --graph     # free check: prints the graph, no model call
python agent.py "Which 5 products had the highest revenue in the last 30 days?"

cd agents_code\01_sql_data_analyst
python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
pip install -r requirements.txt
copy .env.example .env      # then edit .env and paste your API key
python agent.py

Never used Git before? My git and GitHub crash course explains cloning, commits and pushing in Urdu/Hindi.

↑ Back to topNext: Learn LangGraph with the course →

1. Module 00Setup & mental model
2. Module 01Why LangGraph
3. Module 02Your first graph
4. Module 03State schemas
5. Module 04Tools

Each module is written like slides, with at most six bullets per slide, speaker notes, short code snippets, a recap, a 10–15 minute exercise and a list of common mistakes. Open the course folder and start with module 00. If you prefer video, this live session builds a LangGraph agent from scratch:

Live session · 1:16:50

Live session · 1:07:20

Agents are built on large language models, so it helps to know how they work and how to prompt them. Read my prompt engineering guide and the NLP with Hugging Face guide for the foundations.

↑ Back to topNext: Design your own agent with the prompt guide →

1. Pick the closest pattern, not the closest domain: a router, a fan-out, a critique loop, human approval or memory.
2. Fill the master template: agent name, domain, primary user, pattern, hard constraints, tools and data, and tone.
3. Generate the spec with an LLM; it follows the same 11 sections as the example agents.
4. Compare it with the most similar example spec in guide/agents/.
5. Build it by hand or with Claude Code, using the matching agent in agents_code/ as a starting point.

Agent name: Clinic Appointment Assistant
Domain: Healthcare (small private clinic)
Primary user: Patients booking or changing appointments on WhatsApp
Core LangGraph pattern to showcase: router + human-in-the-loop
Hard constraints: no medical advice, no PII sent to the model unscrubbed, staff approve every cancellation
Available tools / data sources: appointment calendar (read/write), clinic FAQ, staff notification
Tone / persona: warm, short, professional; English and Urdu

The full template, with an adaptation checklist, is in guide/00_master_prompt_template.md. To build the spec into an app faster, see how I use Claude Code in my Claude Code guide.

↑ Back to topNext: A learning path: beginner to advanced →

1. Week 1Foundations. Course modules 00–04, then run agents 01 and 02.
2. Week 2Routing and memory. Modules 05–07, then agents 04 and 06.
3. Week 3Parallel and multi-agent. Modules 08–09, then agents 03, 05 and 07.
4. Week 4Production patterns. Modules 10–11, then agents 08–11, and build your own agent with the prompt guide.

Once you are comfortable, see how several agents work together in a real product, and study five more agents with source code:

Live session · 39:39

Live session · 30:04

For even more examples, the companion repository top five AI agents for research and my free book Five Research AI Agents explain research-focused agents step by step.

↑ Back to topNext: Turn the agents into real projects →

• Connect real data through the Tools section, and give each tool only the permissions it needs
• Keep humans in the loop for payments, emails, cancellations and anything risky
• Protect personal data: reuse the PII-scrubbing ideas from agents 05 and 09
• Test before launch with real questions, and log what the agent does

To deploy an agent on your own domain with HTTPS, follow my guide to deploying AI agents on a Hostinger VPS, or package it with Docker. For agents that answer from documents, add retrieval with my RAG tutorial in Python. And to sell these as services, read freelancing with AI and AI agents in the global market.

Live session · 1:02:47

Live session · 21:45

All 10 videos: AI agents and LangGraph

Every lesson used in this guide, in Urdu/Hindi. Visit the Codanics YouTube channel

Links and tools

• 1
  11 AI agents repository (GitHub)Source code, course and prompt guide
• 2
  agents_code folderRunnable code for all 11 agents
• 3
  course folderThe 12-module LangGraph course
• 4
  guide folderMaster prompt template and 11 specs
• 5
  LangGraph documentationOfficial docs for the framework
• 6
  Anthropic ConsoleCreate your API key
• 7
  Claude models overviewCurrent Claude models and prices
• 8
  Top five AI agents for researchMore agents with source code

Related guides on Codanics

• 1
  Python ka Chilla: Build AI AgentsThe course behind these agents
• 2
  RAG tutorial in PythonAdd document search to your agents
• 3
  Deploy AI agents on a VPSPut your agent online with HTTPS
• 4
  Freelancing with AI and AI agentsSell agent projects to global clients
• 5
  Claude Code for content creationBuild faster with Claude Code
• 6
  Five Research AI Agents (free book)Research agents explained with code
• 7
  Prompt engineering guideWrite better system prompts
• 8
  NLP with Hugging FaceLanguage model foundations
• 9
  Docker for AI developersPackage agents for deployment
• 10
  Pandas mastery guideData skills behind the analyst agents
• 11
  VS Code as your IDESet up your editor
• 12
  Free AI & data science roadmapWhere to start if you are new

More AI, data science and career guides on the Codanics blog.

Key terms explained

Short definitions of the words used in this guide

AI agent

A program where a language model decides the next step and uses tools, such as a database query or a search, to finish a task.

LangGraph

An open-source Python library for building agents as graphs of steps (nodes) connected by edges, with state, memory and human approval.

State

The shared data the graph passes from step to step, such as messages, results and counters.

Node

One step in a LangGraph graph, usually a Python function that reads and updates the state.

Conditional edge

A connection that chooses the next node based on the state, for example routing by intent.

Tool calling

Letting the model call Python functions you define and use their results.

Checkpointer

Saves the graph's state so a conversation or paused workflow can continue later.

interrupt()

A LangGraph function that pauses a graph for human input or approval; you resume it with Command(resume=…).

Send (fan-out)

A LangGraph feature that runs the same node many times in parallel, for example once per resume.

Store

LangGraph's long-term memory that survives across sessions, such as a shopper's preferences.

Guardrail

A rule in code that blocks unsafe input or output, such as personal data or medical advice.

PII

Personally identifiable information, such as names, phone numbers and addresses.

Frequently asked questions