Browse by problem

Corrective RAG: grade what you retrieved before you answer

Every retrieved passage is graded for relevance, the failures are discarded, and the answer is written only from what survived. If nothing survives, the pattern refuses to answer rather than inferring from unrelated context.

Dry-Run: simulate before you execute

The agent proposes an action, a simulator predicts its effect and reversibility, and a deterministic policy gate decides whether it runs. The model never reaches the real function; the boundary is an `if` in Python, not an instruction it can talk past.

Episodic and semantic memory: remember turns, and remember facts

Two stores answering different questions. Episodic keeps what was said and when; semantic keeps extracted, deduplicated facts about the user. Retrieval defaults to facts, and reaches for transcripts only when the question is genuinely about what happened.

ReAct: reason, act, observe, repeat

The model alternates between explicit reasoning and tool calls, feeding each observation back until it can answer. The reasoning is a first-class inspectable field rather than something that happened inside the model, and that is what separates ReAct from plain tool-calling.

Introduction

Runnable example (beginner) for script using openai, python-dotenv.

Making Requests

Runnable example (beginner) for script using openai.

Streaming

Runnable example (beginner) for script using openai.

Introduction

Runnable example (beginner) for script using openai.

Level 1: Augmented LLM — Single API Call

One model call with structured output, system prompt, and context. No loops, no tools.

Level 2: Prompt Chains & Routing — Deterministic DAGs

Multiple LLM calls in a fixed sequence. Code controls the flow, not the model.

Level 3: Tool-Calling Agent — Scoped Autonomy

The agent decides which tools to call and in what order, but only within a fixed set of well-defined capabilities.

Level 4: Agent Harness — Full Runtime Access

Give the agent a full runtime via the agent runtime SDK. It can search files, read docs, and reason through problems autonomously.

Basic

Runnable example (beginner) for script using openai.

Structured

Runnable example (beginner) for script using openai, pydantic.

Tools

Runnable example (beginner) for script using openai, pydantic.

Retrieval

Runnable example (beginner) for script using openai, pydantic.

Client Sse

Runnable example (advanced) for script using mcp.

Client Stdio

Runnable example (advanced) for script using mcp.

Client Streamable Http

Runnable example (advanced) for script using mcp.

Server

Runnable example (advanced) for script using mcp, python-dotenv.

Extraction

Runnable example (intermediate) for script using docling, ipykernel.

Chunking

Runnable example (intermediate) for script using docling, ipykernel.

Embedding

Runnable example (intermediate) for script using docling, ipykernel.

Search

Runnable example (intermediate) for script using docling, ipykernel.

Tutorial: Notebook to Web App in Five Minutes

This notebook shows how to turn a small Python function into a shareable interface with Gradio or Streamlit. It focuses on one clear input -> processing -> output flow so the app wrapper stays explicit and easy to review.

Quickstart

Runnable example (beginner) for script using pydantic.

Agents

Runnable example (intermediate) for script.

Dependencies

Runnable example (intermediate) for script using pydantic.

Get Single Page

Runnable example (intermediate) for script using docling, openai.

Web Search

Runnable example (intermediate) for script using docling, openai.

Search Handbook

Search Agent

Runnable example (intermediate) for script using docling, openai.