Docs

Elastic Compute Service

Quickly deploying OpenClaw via application templates

OpenClaw FAQ

This topic summarizes common issues encountered during the deployment of OpenClaw and integration of AI assistants on ECS instances, along with their corresponding solutions.

Type	FAQ
Integrate OpenClaw	How do I quickly integrate OpenClaw on an ECS instance I have already purchased?
Use OpenClaw	An error "404 The model or endpoint xxx does not exist or you do not have access to it." is reported when OpenClaw deployed on an instance calls a model in the Coding Plan. How do I resolve it? How do I open the WebUI after I deploy OpenClaw on an ECS instance? Why does the error "The requested model does not support function calling, please switch to a different model or contact with administrator if you believe this model should support function calling." occur when I make a function call? After OpenClaw is integrated, the error "400 Input length xxxx exceeds the maximum length xxxx." occurs when I chat with the bot. How do I resolve this issue? How do I handle chat lag or situations where an ECS instance becomes unresponsive when I use OpenClaw? Can I directly copy the installation directory of the command-line tool to another computer for use?

Integrate OpenClaw

How do I quickly integrate OpenClaw on an ECS instance I have already purchased?

To quickly deploy OpenClaw on an existing instance, reinstall the instance operating system by creating an OpenClaw application update task. Subsequently, you can integrate AI assistants for IM applications (such as Feishu, WeCom, and DingTalk). For more information, see Configure the model and integrate a Feishu application after ECS instance purchase.

warning

Reinstalling the operating system will clear all data on the system volume of the instance irreversibly. We strongly recommend that you back up system volume data in advance.

Use OpenClaw

An error "404 The model or endpoint xxx does not exist or you do not have access to it." is reported when OpenClaw deployed on an instance calls a model in the Coding Plan. How do I resolve it?

The 404 error indicating "the model does not exist or you do not have access" is usually caused by the following reasons:

Incorrect model activation status

The Coding Plan package must be subscribed to before the relevant models can be used. Log in to the Model activation page and verify whether the model specified in the error message has been activated.

Incorrect model name

Make sure that the used model name is completely accurate. You can obtain the model name by referring to Configure the model name (specified in the configuration file).

Incorrect Base URL configuration

As a Claude-like tool, OpenClaw should use the Base URL compatible with the Anthropic interface protocol:

https://ark.cn-beijing.volces.com/api/coding

For tools compatible with the OpenAI interface, use the following URL:

https://ark.cn-beijing.volces.com/api/coding/v3

Invalid API key used

Check whether the ARK_API_KEY parameter is configured correctly and the configured key has not expired.

tip

If the issue persists after the above checks, we recommend that you refer to the error code documentation for further troubleshooting guidance or contact technical support for assistance.

How do I open the WebUI after I deploy OpenClaw on an ECS instance?

Allow inbound traffic on port 18789 in the security group of the instance.

Select the "Security Group" tab, and click the "Configure rule" button to the right of the target security group (Default is used as an example in this topic).

Select the "Access rules > Inbound rule" tab, and click the "Add rule" button.

Allow access to port 18789 via the TCP protocol.

Confirm the OpenClaw version.

Execute the following command to confirm the currently deployed OpenClaw version.

openclaw --version

Open the WebUI.

Linux
Windows
Remotely access the WebUI from a local PC.
Execute the following command to obtain the OpenClaw authorization token.

openclaw dashboard

warning
Securely keep the returned token. Leakage of the token may lead to serious security risks.

Execute the following command to map port 18789 of the local PC to port 127.0.0.1:18789 of the OpenClaw server.
tip
Make sure that the terminal window where the command is executed remains open during WebUI access.
Adjust the parameters based on your actual situation:
Replace <user_name> with the login username of the instance where the OpenClaw service is deployed.
Replace <public_ip> with the EIP of the instance where the OpenClaw service is deployed.

ssh -N -L 18789:127.0.0.1:18789 <user_name>@<public_ip>

In the browser on the current PC, visit http://127.0.0.1:18789?token=<your_token> to open the OpenClaw WebUI interface.
tip
Replace <your_token> with the OpenClaw authorization token obtained in Step 1.
Open the WebUI on the instance where OpenClaw is deployed.
Execute the following command in PowerShell to obtain the OpenClaw authorization token.
warning
Securely keep the returned token. Leakage of the token may lead to serious security risks.

openclaw dashboard

Subsequently, you can open a browser on the current instance and visit http://127.0.0.1:18789?token=<your_token> to access the OpenClaw WebUI.
tip
Replace <your_token> with the OpenClaw authorization token obtained in the previous step.

Why does the error "The requested model does not support function calling, please switch to a different model or contact with administrator if you believe this model should support function calling." occur when I make a function call?

This issue occurs because the model you are using does not support function calling. We recommend that you switch to a model that supports function calling. Click here to view the list of models supporting function calling.

After OpenClaw is integrated, the error "400 Input length xxxx exceeds the maximum length xxxx." occurs when I chat with the bot. How do I resolve this issue?

During conversational inference, the model incorporates the full context as input parameters into a single inference request to implement session memory. This error occurs and interrupts inference when the context length exceeds the maximum input limit of the model. We recommend that you reset the session regularly to avoid excessively long context:

Clear via the IM dialog box

If you have integrated OpenClaw with IM tools such as Feishu, enter the /new or /reset command in the OpenClaw bot chat window of the IM tool to start a new conversation and clear the context of the previous one.

Clear via the WebUI

If you chat with OpenClaw through its WebUI, open the WebUI, go to the Chat page, and click the "New session" button to start a new conversation and clear the context of the previous one.

How do I handle chat lag or situations where an ECS instance becomes unresponsive when I use OpenClaw?

When you interact with OpenClaw through multiple channels, or perform updates, restarts, log queries, and other operations, OpenClaw may create a new process for each task. If multiple processes compete for instance resources simultaneously, it may result in issues such as chat lag or instance unresponsiveness. In this case, you can:

Optimize memory usage configuration.

Tutorial video

Replay

Play
00:00 / 00:18 Live

00:00

Fullscreen

Cssfullscreen

1x

2x
1.5x
1x
0.75x
0.5x

Click and hold to drag

Procedure

Execute the following command to create a script file.

vim /root/config_swap.sh

Press the i key to enter edit mode, and then copy and paste the following content into the file.

#!/bin/bash
set -e
# If /swapfile already exists on the current system, skip the setting
if [ -f /swapfile ]; then
echo "The swapfile already exists..."
else
fallocate -l 2G /swapfile
# First, modify the file permissions to grant read and write access only to the root user
chmod 600 /swapfile
# Set as swap file
mkswap -f /swapfile
# Finally, enable the swap file
swapon /swapfile
fi
# To enable swap permanently, add the entry after the line for the root partition.
if ! grep -q "^/swapfile" /etc/fstab; then
echo "/swapfile swap swap defaults 0 0" >> /etc/fstab
fi
# Permanently configure swappiness
if grep -q '^vm\.swappiness' /etc/sysctl.conf ;then
sed -i 's/^vm\.swappiness[[:space:]]*=[[:space:]]*[0-9]\+$/vm.swappiness = 20/' /etc/sysctl.conf
else
echo 'vm.swappiness = 20' >> /etc/sysctl.conf
fi
sysctl -w vm.swappiness=20

Press the Esc key to exit edit mode, enter :wq, and then press Enter to save and exit the file.

Execute the following command to run the script file and optimize memory usage configuration.

bash /root/config_swap.sh

If the output is as shown below, it indicates that memory usage optimization is completed.

If the issue persists after you optimize memory usage configuration, we recommend that you:

(Recommended) Upgrade the instance configuration and deploy OpenClaw on a higher-specification instance (such as 4 vCPUs, 8 GiB). Learn more.

Restart the target instance in the ECS console. Learn more.

Can I directly copy the installation directory of the command-line tool to another computer for use?

No. Each time the tool sends a request, the server verifies whether the device fingerprint matches the AppID. If you directly copy the tool installation directory to another computer, the device fingerprint will not match. When you run the tool on that computer, the system will detect the environment change and mark it as unregistered. You need to rerun the installer on that computer to register and associate the new device.

Last updated: 2026.08.19 14:10:34