[Part 1: Basics] OpenClaw.AI Master Class: Create Your Own Autonomous AI Agent with Gemini and Docker

Hello.

In recent versions of OpenClaw since v2026.7.1-beta.5, the onboarding method has changed from the existing menu-type screen to a Crestodian-based interactive onboarding method.

Since it looks different from the original lecture screen, you might find it confusing, so we are providing the following guidance.

Target Audience

- Students who pull and install/practice with the latest Docker image

- Unlike the previous lecture screen, students who see the Crestodian TUI screen first when running openclaw onboard

If you do not fall under the following:

- Students who are still using the old Docker image

It can proceed almost identically to the existing lecture screen, so

Please follow along with the lecture content.

What has changed in the latest version

Previously, the model selection menu often appeared immediately after running openclaw onboard, but

In the latest version, a system agent for setup and repair called Crestodian runs first.

For example, you may see a guidance screen like the one below first.

- "Hi, I'm Crestodian — let's hatch your agent."

- "Say yes and I'll set all of that up now."

- "No usable model provider is configured, so the agent cannot answer yet. Configure a model provider now? Say yes or no."

In other words, in the latest version, there are cases where you must enter yes twice to proceed to the existing model selection flow.

How to proceed with onboarding in the latest version

When running openclaw onboard in the latest version, please proceed as follows.

1. When the first Crestodian screen appears, type yes

When you first see a phrase similar to the following:

- "Hi, I'm Crestodian — let's hatch your agent."

- "Say yes and I'll set all of that up now."

Enter the following into the terminal here.

yes

2. If asked again whether to configure a model provider, enter yes one more time.

Next, a phrase similar to the following may appear.

- "No usable model provider is configured, so the agent cannot answer yet."

- "Configure a model provider now? Say yes or no."

Enter it once more here as follows.

yes

In other words, in the latest version, there are cases where you must type yes → yes twice in a row and press Enter to proceed to the model selection and key entry stage.

3. From this point on, it is the same as the lecture flow.

From then on:

- Select the model to use

- Enter Gemini API key

- Basic agent settings

It continues with a flow such as,

This part is not significantly different from the onboarding purpose explained in the lecture.

## Onboarding Completion Criteria

Once you have finished entering the Gemini API key and completing the settings, you may see a message similar to the one below.

- "Your agent is ready."

- "Model: google/gemini-2.5-flash."

- "Gateway: running at ws://127.0.0.1:18789."

- "Say talk to agent to meet your agent right here, or help for everything I can do."

If you see this screen, it means onboarding has been completed successfully.

In other words:

- Model setup complete

- Gateway connection complete

- Basic agent ready

You can consider it to be in that state.

How should I exit after completing the onboarding?

In the latest version, the Crestodian TUI screen may remain as is even after onboarding is complete.

In this case:

- Ctrl + C

- Press Ctrl + C once more

You can exit the TUI by pressing Ctrl + C twice, as shown.

In the lecture, this part may not appear directly as the screen is based on an older version, but

Students using the latest version can exit by pressing Ctrl + C twice after completing the onboarding.

Summary

Students using the latest version only need to remember the following.

1. Run openclaw onboard

2. When the Crestodian information screen appears, type yes

3. If asked whether to configure the model provider, enter yes again

4. Enter the Gemini API key and proceed with the setup

5. When you see the message "Your agent is ready.", onboarding is complete

6. Finally, press Ctrl + C twice to exit

Existing students / Old version users?

- Students who have not pulled the latest image

- Students who are still using the old version of the Docker image

may be similar to the existing lecture screen, so

You can proceed exactly as shown in the lecture.

Conclusion

To summarize,

In the latest version, the onboarding UI has changed to a Custodian-based interactive method, and accordingly, you can understand that the flow has changed to entering yes twice, proceeding with model selection/key input, and finally pressing Ctrl + C twice to exit.

Students using the latest version, please refer to this notice to proceed.

Hello.

OpenClaw is still undergoing frequent version updates. There are about two changes that would be good for you to know, so I am sharing them with you.

1. Regarding the system configuration files automatically generated in the config folder before onboarding

   1. In our lecture, the config folder was completely empty before OpenClaw onboarding, and the OpenClaw system configuration files were automatically generated once onboarding was complete. However, after the update, even before onboarding, running the docker-compose.yml file just once will cause OpenClaw to generate the basic required system configuration files in the config directory. Even if files generated before onboarding exist, you can proceed with the practice following the sequence shown in the lecture.

2. Regarding changes to the options selected by users during the onboarding stage
   There are some options where the names or other details selected by users during the onboarding stage have been slightly changed or removed, so I will list the changes below.

   1. Section 3 > Lesson 11: Hello OpenClaw First Boot and System Check

      • Changes in the OpenClaw Onboarding Hands-on Practice

        • 4 minutes 40 seconds

          • Config handling configuration option

            • (Existing) Select Update values -> (Changed) Select Review and update

          • How do you want to provide this API Key? configuration option

            • (Changed) This option itself has been removed, so you can ignore it in the lecture video.

          • Default model configuration option

            • (Previous) You could select directly from all models -> (Changed) Select Browse all models, then manually enter the text 'google/gemini-2.5-flash' in the next step

          • How do you want to hatch your bot configuration option

            • (Existing) Select Do this later -> (Change) Select Hatch later

📢 OpenClaw.AI Basic Part 2.5 Course Opening Announcement

Hello, I'm Kevin, the instructor for the OpenClaw.AI Master Class basic course.

I am finally posting an announcement as [Basic Part 2.5] Fundamentals of Multi-Agent Team Design and Role Separation has been released on Inflearn. :)

This Part 2.5 further expands the "AI Assistant in my PC" created in Parts 1 and 2,

This is a lecture focused on growing a single Telegram bot into a small AI team with divided roles.

👉[Go to OpenClaw.AI Basics Part 2.5 Lecture]

What we will build together in Part 2.5

- Using Telegram Forum Groups + Topic-based Routing

We will configure a multi-agent environment that operates two agents, content-planner and content-editor, with a single bot.

- To the /workspace/content/planning / /workspace/content/drafts folders

We will actually create a content workflow where planning deliverables and draft/edited versions are separated.

- Through the scenario of "generating ideas and outlines in the planner room → refining the writing in the editor room," we will complete a realistic multi-agent collaboration routine based on manual copy-pasting.

Especially recommended for these people

- Those who have followed OpenClaw Parts 1 and 2 but are still making a single agent do everything

- Those who want to develop a Telegram bot not just as a "single chatbot," but as an AI team divided into a Planner and a Writer/Editor

- Solo creators, newsletter/blog operators, and YouTubers who want to organize their workflow by delegating content planning, drafting, and refining tasks to role-specific agents.

- Developers, planners, and PMs who are interested in multi-agents but want to start light in their current OpenClaw + Telegram environment instead of using a massive framework

Part 2.5 Core Curriculum at a Glance

- Section 1: For both existing and new students

Part 2.5 Joining Route A (5-minute environment check) / B (Minimum setup onboarding) Guide + Super simple review of OpenClaw UI

- Section 2: Single-agent vs. Multi-agent concepts,

Understanding the structure of how multiple agents are deployed within openclaw.json, and checking the current configuration with openclaw agents list

- Section 3:

- Define content-planner / content-editor agents

- Separate workspaces for /content/planning and /content/drafts

- Telegram forum group & planner-planning / editor-editing topic creation

- Find Group ID / Topic ID → Complete agent routing by topic

- Section 4:

- planner ↔ editor semi-manual collaboration practice (based on human copy-paste)

- Preview of the 4-agent automated collaboration pipeline (Leader/Planner/Editor/QA) to be covered in Part 3

Checklist before taking the course

- It is best if you have the WSL2 + Docker + OpenClaw + Telegram Bot environment used in Parts 1 and 2 ready.

- Even if you haven't taken Parts 1 and 2, I have structured it so that you can set up the minimum environment required for the multi-agent practice by following the “Environment Check Route A / Minimum Setup Onboarding Route B” included in Part 2.5.

- You will need a Gemini API key issued by Google AI Studio.

If you have any questions or get stuck while taking Part 2.5, please feel free to leave them on the Q&A board at any time.

Based on your feedback, we will continue to strengthen the practice examples and troubleshooting guides.

I hope this Part 2.5 serves as an opportunity to expand your OpenClaw environment from "one smart assistant" into a "small AI team with divided roles."

Thank you.

[Notice] OpenClaw Latest Version Bonjour-related Error Guide and Solution

1. Symptom Information

As of today (2026-04-27), after updating the OpenClaw image to the latest version (2026.4.x) via docker pull, an issue where the Gateway container keeps restarting is occurring in some environments.

If the following messages repeat in the openclaw-gateway logs, it corresponds to this issue., thì có nghĩa là bạn đang gặp phải vấn đề này.

[plugins] bonjour: advertised gateway fqdn=... state=announcing

[plugins] bonjour: restarting advertiser (service stuck in announcing ...)

[openclaw] Unhandled promise rejection: CIAO ANNOUNCEMENT CANCELLED

openclaw-gateway exited with code 1 (restarting)

In this case:

- The openclaw-gateway container starts up and immediately shuts down every few seconds,

- Dashboard access and Telegram integration may not function properly.

2. Cause – Bug in the Bonjour (mDNS) advertising process

By default, OpenClaw Gateway uses Bonjour (mDNS) to allow this instance to be automatically discovered (advertise) on the same network.

In some Docker / WSL / network (routers, corporate networks, etc.) environments:

- After the Bonjour advertise status gets stuck in the announcing stage,,

- An error CIAO ANNOUNCEMENT CANCELLED occurs, and

- There is a bug where this exception is not handled properly, causing the Gateway process to fall into a crash loop.

A patch is currently in progress, and until then, you can use the following workaround.

3. Solution – Disable Bonjour feature (Recommended)

The Part 1/Part 2 practice sessions use a structure intended for use on a single local machine only, so Bonjour is not strictly necessary.

Therefore, we recommend disabling Bonjour via environment variables.

3-1. When using docker-compose.yml

Add the following environment variable to the openclaw-gateway service..

services:

  openclaw-gateway:

    image: ghcr.io/openclaw/openclaw:latest

    environment:
      - OPENCLAW_DISABLE_BONJOUR=1

      # Other OPENCLAW related settings...

After making the changes, restart with the following command.

docker compose down

docker compose up -d

By applying this setting:

- The Gateway will no longer attempt Bonjour/mDNS advertisements,

- It will run stably without taking the CIAO ANNOUNCEMENT CANCELLED error path.

- It does not affect the browser access (dashboard) and Telegram integration used in Part 1/Part 2.

4. Future Guidance

- This issue is a compatibility problem between OpenClaw's Bonjour plugin and certain network/Docker environments, and it is expected to be improved in a future release.

- Even after the patch,

- If you are using it on only one local machine, you can continue to keep OPENCLAW_DISABLE_BONJOUR=1, and

- You only need to re-enable Bonjour when you need to automatically discover OpenClaw instances across multiple devices.

For those students in Part 1/Part 2 who are experiencing a sudden issue where the Gateway keeps shutting down after recently re-pulling the image,

Please apply the settings above to disable the Bonjour feature.

Thank you!

📢 Notice: OpenClaw.AI Basic Part 2 Course Now Open

Hello, I am Kevin, the instructor for the OpenClaw.AI basic course.

I am writing to inform you that [Basic Part 2] OpenClaw.AI Master Class: A Smartphone Remote Assistant Completed with Telegram and Vision AI has been released on Inflearn. ^^

I hope my lecture can be of even a little help in allowing you to utilize OpenClaw smartly in your daily life.

👉[Go to OpenClaw.AI Basics Part 2 Lecture]

What we will create together in Part 2

- Build an environment to remotely control your agent anytime, anywhere using Telegram on your smartphone.

- Complete a household account book routine where a single receipt photo is automatically analyzed and recorded using Gemini 2.5 Flash's vision capabilities.

- Upgrade the Docker + workspace environment created in Part 1 into a real-world automation scenario.

Especially recommended for these people

- Those who say, "Now I want to use agents on my smartphone without a browser."

- Those who want AI to automatically classify and record receipt/image data instead of manually organizing it every time.

- Those who want to connect OpenClaw, which they only got a taste of in Part 1, to actual work/life automation.

Part 2 Core Curriculum at a Glance

- Section 1: Part 1 environment check and "Agent in My Pocket" scenario orientation.

- Section 2: Creating a Telegram bot with BotFather → Pairing → Real-time connection with OpenClaw agent.

- Section 3: Send receipt photos via Telegram → Automatic extraction of date, business name, amount, and items → Automation of cumulative storage in the expenses/ folder.

Pre-course Checklist

- It is best if you have the Docker container and Gemini API key used in Part 1 ready.

- Even if you haven't taken Part 1, you can set up the minimum required environment by following the “Hands-on Environment Pre-check List” provided in Part 2.

If you have any questions or get stuck while taking Part 2, please feel free to leave them on the Q&A board at any time.

Based on student feedback, I will continue to reinforce the practice exercises and troubleshooting guides.

Hello, OpenClaw Part 1 students.

It seems that many people are getting stuck during the environment setup stage involving Docker, WSL2, macOS, etc., so I have created new full setup guide documents for each operating system separately from the main lecture content.

The following two documents have been newly added.

• Windows (WSL2) Dedicated Setup Guide

  • Install Docker Desktop

  • Installing WSL2 + Ubuntu and integrating with Docker

  • OpenClaw project folder / .env / docker-compose.yml / openclaw.json configuration

  • Connecting Gemini via onboarding

  • Accessing the dashboard → Pairing the device → First greeting in the chat window

• macOS Exclusive Setup Guide

  • Install Docker Desktop for Mac

  • OpenClaw project folder / .env / docker-compose.yml / openclaw.json settings

  • Connecting Gemini via Onboarding

  • From dashboard access → device pairing → to the first greeting in the chat window

Both documents aim for “a state where the first greeting message returns in the chat window without errors,” and include a compilation of frequently encountered error messages and checklists during the setup process.

👉How to check

• You can use the guide document corresponding to your OS via the link below.

  • Go to OpenClaw Full Setup Guide by OS

From now on

1. First, please follow the OS-specific setup guide thoroughly, and

2. If there are still parts that remain unresolved,

   • Current environment (Windows / macOS),

   • the command you executed,

   • If you provide the error message/screenshot
     , we can assist you more quickly.

If there are any changes, the setup documentation will be continuously supplemented and updated,
so if you get stuck at any point, please refer back to the newly uploaded guides.

Thank you!

Hello, I'm Kevin.

I have organized troubleshooting documentation for two issues that frequently occur in Windows + WSL2 environments and added them to the lecture materials.

• When the error The command 'docker' could not be found in this WSL 2 distro occurs in WSL2 Ubuntu

• If you get an "API Key Missing" error when leaving google_api_key / GEMINI_API_KEY blank in auth-profiles.json

If you have experienced the above situation even once, reading through the following document in the lecture materials will help you quickly resolve the same issue on your own if it occurs again.

docs/troubleshooting/03-openclaw-docker and Environment Variable Recognition Troubleshooting Guide.pdf

In the future, I will continue to document and share common points of confusion or frequently asked questions during the practice sessions in this format.

Thank you!

To all students of OpenClaw Basics Part 1,

Hello, I'm Kevin.

In response to your feedback requesting more systematic management of the lecture materials, we have completely reorganized the structure of the lecture materials for OpenClaw Basics Part 1 to make practice and code reproduction much easier!

Thank you so, so much for your feedback.

With this reorganization, we focused on minimizing the "confusion over what to download and from where" as much as possible, so please refer to the details below!

What has changed?

Now, the GitHub repository for downloading lecture materials is organized into the following major folder structure.

• docs/
  You can find documents corresponding to the [Free Guide] in one place, including the installation checklist, Docker/Linux cheat sheets, system prompt templates, and file auto-classification prompts.
  

• lesson-files/
  Contains the initial set of files needed right before starting each lesson.
  Open the folder corresponding to the lesson number (Section/Lesson), copy it to your working directory, and begin your practice.

• releases/
  This is a space where the
  **"step-by-step completed versions" of files such as docker-compose.yml and openclaw.json** are collected as they are developed throughout the course.
  You can use these for reference when you want to "jump" to a specific stage if you run into issues during the practice.

• samples/
  A folder containing answer keys and reference sample data, such as example inputs/outputs, reference IDENTITY/USER examples, and report examples.
  

• slides/
  The slide PDFs used in each lesson are organized here.
  The filenames follow the format basic-p1-sectionXX-lessonYY.pdf, so they match directly with the section/lesson numbers on Inflearn.

• templates/
openclaw.min.json, openclaw.json contains reusable default configuration templates such as templates, system prompts, and user profile templates.
  

How do I find the practice files for each lesson?

1. Check the section/lesson number you are currently taking on Inflearn.

2. Find the lesson-files/sectionXX-lessonYY/ folder on GitHub.

3. Read the README.md in the folder first, and copy the files to your working folder as instructed.

4. Proceed with the practice by modifying and executing the code as explained in the video.

5. If you need to check if everything is correct, you can compare it with the files in the releases/ folder at the same stage.

Where can I find the free guides, prompts, and slides?

• Free Guides (Checklists, Cheat Sheets, Prompt Collections)
  → These are organized in PDF/text format within the docs/ folder.

• IDENTITY / USER Profile Chat Prompts
  → These are also provided as text files within the lesson-files/sectionXX-lessonYY/ folder of the relevant lesson.
  (A copy-paste version has been added so you don't have to manually type out content that only appears in the slides.)

• Slide PDFs for each lesson
  → You can open the basic-p1-sectionXX-lessonYY.pdf files in the slides/ folder.

How should you use this moving forward?

The recommended flow is as follows.

1. Watch Inflearn videos

2. Prepare the initial set of files from the lesson-files of the corresponding lesson

3. Refer to guides/cheat sheets/prompts in docs if necessary

4. Compare the results with releases after the practice session

5. For additional practice, utilize the samples data.

This reorganization is based on actual student feedback.

“I wish the code provision was just a bit more systematic.”

In response to that, we focused on solving the problem of “having plenty of materials but being confused about where to start.”
If you have any inconveniences or think, “It would be great to have this kind of material,” please let us know anytime through questions or reviews.
We will actively reflect them in our updates.

Thank you.
Best regards, Kevin, the operator of [Basics Part 1] OpenClaw.ai Master Class.