NirDiamant/agents-towards-production

A collection of hands-on, code-first tutorials for building and deploying production-grade GenAI agents.

What it is

Agents Towards Production is an open-source set of runnable tutorials covering the building blocks of a production GenAI agent stack: orchestration, memory, retrieval, deployment, security, observability, evaluation, and UI. Each tutorial lives in its own folder with notebooks or code files and, in many cases, its own requirements.txt. It is aimed at developers who want to move agents from prototype to production using concrete, adaptable examples rather than prose documentation.

Key features

A stated 28+ production-grade tutorials, each self-contained and runnable.
Tool integration examples including Model Context Protocol (MCP) and secure OAuth2 tool calling (Arcade).
Memory tutorials using Redis (dual-memory and semantic search), Mem0 (hybrid vector and graph), and Cognee.
Deployment paths covering Docker containerization, on-prem LLMs with Ollama, AWS Bedrock AgentCore, and GPU scaling on RunPod.
Frameworks and serving via LangGraph, FastAPI endpoints, and a Kotlin agent with JetBrains Koog.
Security and evaluation content, including LlamaFirewall guardrails and hands-on prompt-injection testing.
Many tutorials are contributed by named sponsor companies, each demonstrating a specific tool.

Tech stack

Primarily Jupyter Notebooks, with Python as the main language.
Dependencies are declared per tutorial via requirements.txt; there is no root manifest.
Referenced tools and frameworks across tutorials: LangGraph/LangChain, FastAPI, Docker, Ollama, Redis, Mem0, Cognee, MCP, and the A2A protocol.
At least one tutorial uses Kotlin (Koog) rather than Python.

When to reach for it

You want to add a specific production concern (memory, observability, security, deployment) to an existing agent.
You prefer runnable, copy-and-adapt examples over conceptual documentation.
You are evaluating a particular vendor tool in the context of an agent workflow.
You are learning the end-to-end path from a prototype agent to a deployed one.

When not to reach for it

You want a library or framework to import — this is educational content, not an installable package.
You want vendor-neutral coverage — many tutorials are sponsor-contributed and centered on a specific product.
You need a maintained runtime dependency — the material is meant to be studied and adapted into your own code.

Maturity signal

Created in mid-2025, the repository has around 21,000 stars, a last push in July 2026, and only 12 open issues, which fits an actively curated tutorial collection that receives steady additions rather than bug-driven churn. The license is reported as NOASSERTION, meaning GitHub did not detect a standard license, so the LICENSE file should be reviewed before any reuse. Overall this reads as an actively maintained, high-traffic educational resource.

Alternatives

The LangChain / LangGraph documentation and templates — use them when you want the framework's own first-party reference material.
The OpenAI Cookbook or Anthropic cookbook — use them when you want provider-authored example code.
An awesome-list style index of agent resources — use it when you want a link directory rather than runnable tutorials.

Notes

The license is NOASSERTION, unusual for a repository of this size, so the exact reuse terms are not machine-detected.
The README is heavy on sponsor logos, and a large share of the tutorials are contributed by the sponsoring companies to showcase their own platforms.
Although the repository is Python-notebook centric, it spans multiple runtimes, including a Kotlin (Koog) tutorial.

Wiki: NirDiamant/agents-towards-production