LongCut logo

【自动AI】自动化AI员工!NanoClaw+OpenClaw免费打工,你只管摸鱼 | NanoClaw+OpenClaw:FreeAI Workers While You Just GoofOff🚀

By 科技梦星云

Summary

Topics Covered

  • Containers Isolate AI Safely
  • Mobile Apps Control AI Agents
  • Free APIs Often Fail
  • Paid Keys Unlock Reliability

Full Transcript

Hi everyone, I'm Mo Duan.

Welcome to EdgeTechNebulae.

Today we're discussing two "lobster-related" tools.

One is NanoClaw, aka the "Little Lobster".

The other is OpenClaw.

We'll focus on the free API options, meaning no-API or free API alternatives.

Later, we'll try deploying to the cloud.

And eventually, local offline deployment, which might allow for a completely free setup.

But for now, let's look at free API options.

NanoClaw, the "Little Lobster", is like "Feifei" from the Classic of Mountains and Seas: Light, compact, docile, isolated, always ready, and smart.

OpenClaw is like the "Xuanguiyin": Shell holds all, memories of millennia, one step offline, Secure and stable for ages.

The main difference is Xuanguiyin (OpenClaw) doesn't strictly require an Anthropic key.

Let's go step by step.

First, let's look at NanoClaw.

What does this "Little Lobster" do?

It's a miniature version of OpenClaw.

Basically similar functions, but with one great feature: Isolation.

It doesn't run directly on your OS.

If installed locally, it runs inside a container.

On a Mac, it likely uses Apple Container.

On other systems, you can use Docker.

Running it in a container makes it much safer instantly.

To translate: This makes it clearer.

Commands run inside the container, not on your host machine.

So bash access is secure.

It's also very compact.

Simply put, the code is concise.

It lets AI solve problems for you.

Note that it runs on the ClawAgent SDK, so ClawCode is required.

Here's the quick start guide.

These are the install commands.

This is setup.

Currently, you can use WhatsApp to message NanoClaw.

Meaning you can actually control it via mobile apps.

It's also customizable.

Here's the custom command.

To use the Feifei NanoClaw, there are prerequisites.

ClawCode is required, as is a container.

Finally, some FAQs.

You can find answers here.

It uses an MIT License, so it's free to use.

Lots of stars, as seen at the top.

Over 1000 stars already.

Give it a star.

There's no Release version, so you must install via CLI.

Okay, let's follow the steps one by one.

First, use these two commands to pull NanoClaw down, then enter its directory.

Press Enter.

Inside, use this command to check if everything is pulled.

Then install ClawCode with this command.

As mentioned on the site, ClawCode is mandatory.

Check the version after install to ensure it's correct.

Enter to start installing.

It installs quickly.

Version is 2.1.37.

Next step per the site is to run this command.

This handles dependencies, auth, container setup, and configs.

Basically, a setup command.

Enter the setup command in terminal.

Command not found?

Let's see why.

Need to run NanoClaw first.

Use this command to run it.

Then start setup. 99 00:19:00,080 --> 00:04:20,480 We use this command to start NanoClaw.

Seems started.

After the start command, you choose a mode.

Pick one you like.

Or use default if unsure.

I'll use default.

Now you need a login method.

Basically, it needs an Anthropic account.

If you have an account, you'd best have an API key.

You can pay a small amount to buy a key to test.

No need to invest much yet.

Test its capability first.

If you're a subscriber with a Pro account, choose the first option.

If you just bought a key to play, choose the second.

If using other accounts like Amazon Bedrock, Microsoft Azure, or Vertex AI, you can login too.

Regardless, a login account is required.

Pick according to your preference.

Once selected, it logs in via web.

Usable after login.

We'll return after I login.

Press Enter to continue after login.

Some security warnings appear.

Ignore and press Enter.

Next is Terminal settings.

We'll use the default recommendation.

Press Enter.

Select current directory.

Press Enter.

Now, we can finally enter commands.

Like entering the "setup" command.

Give it a Setup command.

Select using arrow keys.

Enter to confirm.

It starts helping us setup.

Click Continue to proceed.

You can see on my local machine, Apple Container is installed.

So I don't need to install it.

But you can also use Docker.

If you prefer Docker, that works too.

Open another port.

How do we know Apple Container is installed?

If you're on Mac, use this command to check version.

If it's higher than 13, it's installed.

If not, it's missing.

Go to Apple Container's official GitHub.

Here is the project.

Click into Releases.

Here's the installer.

Install it if you need.

I have it installed.

Let's press Enter.

Looks like I was wrong.

Apple Container is NOT installed.

I must have lost it during the OS update.

My bad.

But it offers an option to install Apple Container directly for you.

If you are on a Mac, press Enter.

Here are brief install/start instructions.

Let's follow them.

Download the PKG.

Run it after download.

Double click.

Don't "Move to Bin", click "Open".

Go to Privacy settings, allow it to run.

Then double click again.

It should run now.

I'll set Privacy and come back.

Allowed "Open anyway".

Double click to run.

Click Continue.

Click Install.

Enter credentials, install software.

Installing Container.

And it's done.

Close.

Now you can Move to Bin.

Next: `container system start` to start the service.

The official docs say the same.

Open a new terminal.

Input `container system start`, Enter.

It needs to install some layers.

Press Y to agree.

This might take a moment, it's quite big (456MB).

456MB.

But won't take too long.

Installed.

Use the command again.

Judging by the feedback, the container is running.

Back to NanoClaw.

Enter this line.

See if it detects 217 00:08:58,080 --> 00:08:57,880 Apple Container is installed/running.

It pops up saying Apple Container is installed and running.

Next, enter API Key.

Pick your preferred one.

It recommends subscription.

Your choice.

Entering API Key needs an env file.

Configure one or input directly.

Direct input restarts the command prompt.

Better to let it create a `.env` file.

Continue? Yes.

Input API Key here.

I'll input mine, you yours.

Enter when done.

Key entered, continue? Yes.

continue? Yes.

Verify API Key? Yes.

API Key configured.

Now configure a Nano Agent.

Continue? Yes.

Wait a bit, configuring Nano Agent takes time.

After config, test Container Runtime.

Select Yes, wait again.

Now authorize WhatsApp.

Command is `npm run auth`, select Yes.

There's a QR code, but it looks incomplete/glitchy.

Can't see clearly.

Generate another one.

Instructions on connecting WhatsApp appear.

Ready to authorize again.

Select Yes, image still unclear.

Open a new terminal.

Run `npm run auth` directly.

Now we have a complete QR code.

Use WhatsApp to link.

Scan this big QR Code.

And link.

Wait, "Connection Failed".

Running verification again, it asks to delete auth and retry.

Let's delete it.

Use the command again.

See if it generates a QR Code.

QR Code generated.

Still says link failed.

Ignore it, back to Nano.

Choose an Assistant role.

Select Custom Name.

To define a name.

That's the idea.

Naming it "Hello World".

Now it's "Hello World".

Then about how you communicate via WhatsApp.

Select "Recommended".

We'll choose that.

Seems WhatsApp really didn't connect.

Enter phone number here.

To get a code?

I'll input it off-screen.

After inputting number, it asks to Build.

Then asks to `run dev`.

Running locally.

Select Yes.

Verification starts again.

Authorize WhatsApp. Yes.

Maybe because it was authorized before, but failed.

So it asks to delete and re-authorize.

Select Yes.

Starting authorization again.

Select Yes.

Glitchy image again.

Open a new terminal.

Run `npm run auth` ourselves.

Big QR code appears.

Scanned but still can't connect.

Might be my network connection is unstable.

Don't give up.

Authorize again.

Success message appears.

WhatsApp authorized successfully.

Back to Nano.

It asks for verification again.

Let it confirm.

Is it authorized?

Nice.

It reads the auth info, says authorized.

Now Build.

Select Yes.

Test WhatsApp connection.

Select Yes.

Create conversation document.

Select Yes.

Create group files?

Select Yes.

Save conversation to memory/cloud.

For long-term memory settings.

Select Yes.

Editable?

Select Yes.

External access?

As recommended, No.

Create Allowlist?

Select Yes.

Continue?

Yes.

Configure Launchd Service.

For auto-run or background running.

Yes.

Node needs filesystem access.

Agree.

Verify NanoClaw is running.

Agree.

Running the command.

Checking status.

Now check stopping service and retrying.

Yes.

Authorize WhatsApp again. Yes.

Glitchy icon again.

Open a new terminal.

Run the auth command manually.

Connection failed again.

Run it again.

Now it says success.

It says scanning must be within 20s.

Let me authorize again.

Re-authorization command.

Yes.

Continue? Yes.

Still can't connect.

Run again.

Check if success.

Says successful.

Might be due to different terminal windows.

Causing this issue.

One failed, but the other succeeded.

Let it confirm.

Maybe it worked.

Popup says success.

Now start NanoClaw service.

Yes Agree.

Trigger is "Hello World", the bot name we set.

Looks working.

Ready to chat.

I'll say "Hello" and see the reaction.

It responded in WhatsApp.

Find "Message Yourself" in WhatsApp.

That's what we set up.

Now we can ask questions, or ask it to do things.

See if it opens Chrome.

Open Google, click Send.

It says opened, but I don't see it.

Let's ask for a screenshot.

See if it works.

Checking for screenshot capability and automation features.

Or if it's mostly for search.

Currently, screenshots seem half-baked.

Might need "skills" config to fully work.

Let's test online search capabilities.

Why did WhatsApp auth fail repeatedly?

Does it support Chinese?

Send.

It supports Chinese.

Reading info.

Good.

It summarized it.

Auth failure is due to the 20-60s timeout.

It expires.

So within that window, you must authorize.

It supports Chinese, and can debug itself.

Check the trends of these two tools.

Send.

Quick summary.

Nice.

Sources cited below.

Seems reliable.

Explore more fun ways yourself!

NanoClaw's biggest pro is secure isolation.

It runs inside a container, protecting your host.

Played the small one, now the big one.

OpenClaw.

I demonstrated it last time.

Installation and deployment worked, but the OpenAI API key quota was exceeded, so Telegram didn't respond.

As seen earlier, a valid key works perfectly in NanoClaw.

So it would work in OpenClaw too.

Definitely.

But the point is OpenClaw has free options, or no-API-key options.

I want to try them.

Ideally free.

It has nearly 180k stars.

Spending money is easy, just top up a key.

But my viewers know I focus on free solutions.

Let's go.

Install following the official guide.

We're testing free API key options.

Copy paste this command.

Install OpenClaw.

Detailed steps are in the previous video.

Check it out if interested.

We're mainly testing the API key part.

So most settings will be default.

Run this command.

Enter.

Accept risks.

Quick start.

Many API key options here.

Some are free.

Synthetic is simulated, not real AI.

No key needed, but results might be unsatisfactory.

This OpenAI option needs a key, but has a short free trial.

Of course, it's unstable.

OpenRouter and Google are also free.

Both OpenRouter and Google offer free keys.

Between them, I'd pick Google Gemini.

But OpenRouter has many models.

Quick tip: How to get its key and model.

If you want Qwen, search Qwen.

Free models appear.

Register an account for the key, click account -> keys.

Create key there.

We're demoing Synthetic and Google.

Trying no-key Synthetic AI first.

Enter.

Slap in the face.

Synthetic NEEDS a key.

Select Back.

Hoping for Google Gemini now.

Select Google.

Select Gemini API key.

If you need a key, go to Google AI Studio.

Dashboard -> API keys.

Click this.

Create a key.

Copy and save it securely.

Somewhere accessible.

Click this.

Select Gemini API key.

Paste key.

Enter.

Many models here.

We'll pick the default.

Enter.

Mobile communication?

Telegram again.

Since we used WhatsApp for NanoClaw.

Telegram here.

Brief intro.

This is old.

Don't need it.

Delete.

Find Bot Father.

Slash.

Select NewBot.

Like the guide says.

Give it a name.

And again.

Get the Token.

Copy Token.

Paste here.

Enter.

No Skills needed yet.

Installing Gateway.

Incubating robot.

Select recommendation.

It seems running.

Check the official documentation.

Select Getting Started.

Since Gateway is ready, check Gateway status.

If okay, open Dashboard.

Copy command.

Paste failed.

Can't open.

Checking problem.

Useless command here.

No message response.

Maybe it's the TUI interface.

Exit.

Ctrl+C to exit.

Clear page.

Copy command.

Paste.

Gateway connected.

Open UI with this command.

Enter.

Looks connected.

Status: Connected.

Health: OK.

Click Chat.

Messages were already sent.

That explains it.

Maybe send here?

Probably useless.

Click Send.

Still no response.

I gave the API Key.

Why no response?

Use this command.

Pair OpenClaw and Telegram.

Pairing successful.

So free Gemini keys don't work.

Delete bot.

Delete.

Useless.

Moving on.

Last one failed.

Delete previous OpenClaw.

Trying Anthropic API Key.

Should work.

Install directly.

Copy command.

Paste.

Enter.

Start config.

Confirm risks.

Quick start.

Select Anthropic.

Select API Key.

No API Key yet.

Paste API Key here.

Enter.

Select Model.

Many models here.

Pick one you like.

Telegram chat.

Select Bot Father.

New one.

Name it.

Name again.

Same name actually.

Second is User.

Copy Token.

Paste.

Enter.

No Skills needed.

Install Gateway.

Gateway ready.

Incubate robot.

Recommend option.

Wait.

Almost done.

Looks good.

Assign a role.

Name it.

Check if it works.

Wow, it reacts!

Paid keys just work.

Ctrl+C to exit.

Connect to Telegram properly.

For mobile access anywhere anytime.

Open new bot.

Click Start.

Pairing code.

Command + Code to pair Telegram.

Telegram seems paired.

Send a message: "Who are you?"

Waiting for response.

The dots indicate response.

Dots flashing.

It responded!

Great.

Money well spent.

Open an empty page?

Needs more actions to execute.

Ask about its ability to gather real-time info and Chinese support.

Understood.

Needs "Skills" config (corresponding services) to perform tasks.

That's its purpose: A massive automation butler.

You configure the functions.

With Anthropic API Key, it runs completely and generates replies.

Free options might be unusable.

As stated on their site.

Covered two tools: Feifei NanoClaw, and OpenClaw.

Though OpenClaw needed a paid key.

Give it a try!

Please Like & Subscribe to support me.

Thanks.

Loading...

Loading video analysis...