euleragent Basic Tutorial Collection

This directory contains 9 step-by-step tutorials for developers new to euleragent. Each tutorial can be read independently, but following them in numerical order is most effective.

Tutorial Index

Recommended Learning Order

The recommended learning path is as follows. Each tutorial builds upon the previous one.

01 → 02 → 03 → 04 → 05 → 06 → 07 → 08 → 09
Getting  Modes  Approval  Batch  Memory  Dynamic  RAG   MCP   LLM Profiles
Started

If you are installing euleragent for the first time, be sure to start with Tutorial 01. Workspace initialization and agent creation must be completed before the subsequent tutorials will work smoothly.

Tutorial Descriptions and Learning Outcomes

01. Getting Started — Workspace Setup and First Run

Experience the complete basic flow of euleragent in one go. This covers the end-to-end hands-on process from installation to first agent run, approval handling, and audit log review.

Learning Outcomes: - Install euleragent and initialize a workspace - Create an agent (euleragent new) - Run your first task and review results - Check the pending approval list and batch accept - Read the audit log

02. Plan Mode vs Execute Mode

Run the same task in two different modes and compare the results directly. Practice with the agentic loop behavior and the --max-loops, --min-proposals, and --artifact-name options.

Learning Outcomes: - Understand the differences between Plan mode and Execute mode - Learn criteria for deciding which mode to use - Control the agentic loop with --max-loops - Specify custom artifact filenames

03. HITL Approval Workflow

Fully understand the HITL (Human-In-The-Loop) approval flow where humans review and control agent actions. Practice with approval record structure, parameter modification, and rejection handling.

Learning Outcomes: - Understand the approval record JSON structure - Accept, modify, or reject individual approvals - Apply handling policies by risk level (low/medium/high) - Use batch approval filtering (--tool, --run-id)

04. Task Files, Variable Substitution, Batch, and Resume

Learn how to manage complex research tasks as files, make them reusable with variables, and resume from previous runs.

Learning Outcomes: - Best practices for writing markdown task files - Variable substitution with --var KEY=VALUE - Enforce batch information gathering with --min-proposals N - Plan-to-execute handoff with --resume-run

05. Chat Mode and Long-Term Memory

Learn the interactive chat mode and practice how the agent stores user preferences in long-term memory, remembering them across sessions.

Learning Outcomes: - Interactive conversation with euleragent chat - Teach preferences, prohibitions, and habits, then verify memory - Resume previous sessions with --continue - Query memory status and perform manual searches

06. Dynamic Workflow

Learn the --dynamic flag that enables the LLM to automatically decompose complex tasks into multiple phases and execute them in order.

Learning Outcomes: - Understand what --dynamic does internally - Check workflow status with workflow show - Execute and resume phase by phase with workflow resume - Integrate with external systems via REST API

07. Web Search + Local Knowledge Base (RAG)

Learn how to use the web.search tool's HITL security gate together with local document-based RAG (Retrieval-Augmented Generation).

Learning Outcomes: - Use rag init/add/status/query commands - Add local documents to the knowledge base - Configure web search security (domain whitelist) - Real-world scenarios combining web search and local KB

08. MCP Provider Setup and Tool Management

Learn how to connect MCP (Model Context Protocol) servers to integrate various external search sources (Tavily, Brave, custom APIs, etc.) through the unified web.search interface.

Learning Outcomes: - Understand the MCP protocol concept and its role in euleragent - Configure and connect MCP servers in workspace.yaml - Manage catalogs with mcp sync / mcp show - Handle the source_enable approval flow - Understand catalog snapshots and fail-closed security policy - Configure multiple MCP servers with different trust levels

09. Scoped LLM Profiles and External LLM Approval

Learn multi-profile configuration using different LLMs for the planning stage and the output generation stage, along with the complete HITL approval workflow cycle required when using external LLMs.

Learning Outcomes: - Manage multiple LLM profiles with llm_profiles configuration - Select scoped profiles with --llm-plan / --llm-final - Experience the external profile fallback-to-approval-to-rerun cycle - Modify profiles during approval with --edit-params - Batch approve with accept-all --tool llm.external_call - Understand scope application in Dynamic mode

Prerequisites

Please prepare the following items before starting the tutorials.

Required Environment:

Testing Without Ollama:

To proceed with the tutorials without an actual LLM, set the following in .euleragent/config/workspace.yaml:

default_llm_profile: fake

The fake provider returns deterministic dummy responses, allowing you to test the entire flow without a network connection.

Verifying Installation:

pip install -e .
euleragent --help
euleragent doctor

If the doctor command returns OK for all items, you are ready to go.