Tutorials#

These tutorials document the v1-compatible Python façade retained in OpenRath v2.0.0. They are organized in the order you are likely to use it: build a Session, understand where tools run, try complete Colab projects, then move into the numbered example ladder.

Each tutorial focuses on code snippets, key-line notes, and observed behavior. Together they cover the common path from basic API usage to Workflow customization.

Core Learning Path#

Order

Page

What it covers

1

Session Basics

Create user and agent sessions, and understand fork(), detach(), and Backend placement.

2

Local Sandbox Tools

Open a local Backend directly and see how file, command, and code payloads run around a workspace.

3

Session Loop Tool Calls

Understand model tool calls, tool dispatch, tool_result chunks, and the next completion round.

4

Custom FlowToolCall

Define your own tool schema and Python execution logic, then pass it into the Session loop.

5

End-to-End Colab Examples

Build a literature review, an editable presentation, and an Excel dashboard with durable OpenRath v2.0.0 workflows.

6

Runnable Examples

Move into the v1-compatible 01-12 learning ladder included in v2.0.0, including memory, dynamic routing, and workflow compile.

End-to-End Colab Examples#

These three projects come from the public Rath-Team/example repository. Each Notebook uses fixed inputs and support code, runs an OpenRath v2.0.0 durable workflow, verifies the result, and offers the final artifact for download.

Example

Durable workflow focus

Final artifact

Colab 01: Literature Research

Checkpoint recovery, source constraints, and an evaluation gate.

literature-review.md

Colab 02: PPT Creation

A persistent approval interrupt followed by resumable delivery.

openrath-presentation.pptx

Colab 03: Excel Visualization

Data-quality routing, formula audit, and worksheet previews.

sales-dashboard.xlsx

Example Groups#

Group

Use it for

Entry

Basics

See the smallest runnable paths through agents and lineage.

01 Hello Agent, 02 Session Lineage

Runtime & Tools

Check where tools execute and how tool calls are represented.

03 Sandbox Backend, 04 Built-in Tools, 05 Custom Tool

Integrations

Try MCP tools, streaming, compression, memory, and provider variation.

06 MCP Tool, 07 Streaming, 08 Compress, 09 Memory, 10 Provider Variation

Orchestration

Route dynamically, then inspect the static resources that can be known before execution.

11 Dynamic Selector, 12 Workflow Compile

Choose by Task#

Task

Start with

Understand OpenRath’s state model

Session Basics

Check which directory tools run in

Local Sandbox Tools

See how an agent calls tools across turns

Session Loop Tool Calls

Wrap an external API as a model-callable tool

Custom FlowToolCall

Connect OpenSandbox

03 Sandbox Backend

Try a streaming UI callback

07 Streaming

Use an Anthropic model

10 Provider Variation

Wrap an MCP server as tools

06 MCP Tool

Inspect session lineage

02 Session Lineage

Try key-free local memory

09 Memory

Route dynamically between workflows

11 Dynamic Selector

Inspect workflow resources before a run

12 Workflow Compile

How to Read#

Each page uses the same structure:

  1. Read the coverage table first to confirm what the page explains.

  2. Follow the code steps to understand the API boundary.

  3. Compare the key-line notes to see where state changes.

  4. Run or rewrite the exercises to turn the example into your own code.

  5. If behavior is unexpected, check the troubleshooting table first, then use Developer Notes for source-level details.

Tutorials