OpenClaw Lobster (formerly ClawdBot/MoltBot)

OpenClaw Lobster (formerly ClawdBot/MoltBot)

1. Recommended cloud servers (ignore this if installing locally)

Cloud deployment is strongly recommended: 1. Online 24/7 2. A cloud environment makes deployment cleaner and simpler 3. Overseas servers can access the open internet directly, so you don’t need to buy local proxy/VPN software 3. Not tied to your local computer, so you can use Lobster from any computer or phone

Alibaba Cloud: The advantage is heavy optimization, fast speed, and strong performance. A 2C2G server can match the performance of some other providers’ 4C4G servers.

Tencent Cloud: The advantage is a simplified, visual page for configuring large models, and the network speed is still pretty good (the tutorial for connecting to their own QQ is very thorough).

Although running Lobster on a cloud server gives you the advantage of 24/7 uptime, if you are a developer, we strongly recommend installing one locally as well: use your local machine for exploration, and use the cloud server for stable, real-time business work. If you have no technical background, we recommend buying a dedicated cloud server or a Mac mini for this purpose, to avoid permanent damage to your work computer since Lobster has the highest system privileges.

---

2. Configuration and usage (choose one based on your scenario; don’t be afraid!)

Option 1: (Strongly recommended locally) Youdao Lobster configuration

For local installation, we recommend Youdao Lobster first. It supports Windows and Mac. Click here to download

At present, aside from the official original installation, the many “Lobster shell” products on the market are a mixed bag, and many are small-shop builds with profit-driven behavior or backdoors. If you’re installing on a local computer, we recommend Youdao Lobster. This is the simplest and most reliable method in my opinion!!! (Trust Lei Zi! Trust NetEase!)

In the image below: 1. Set URL and key 2. Add model 3. Connectivity test 4. Start using it

Connecting to Feishu, DingTalk, QQ, and other bots

The interface is very simple. Just create the ID and secret on each platform. If you don’t understand something, just ask any AI to help you and it will be configured successfully.

---

Option 2: (Not recommended for now!) Configure in the onboard wizard

This has been verified: models added this way only get a default context window of 4096 from the official settings (apparently the official team doesn’t know what they were thinking; even qwen3-0.7b should not be this small). Our mainstream models now have at least 200K context, and I set mine to 256000. 4096 is far too small — that’s only a few thousand Chinese characters, so you can’t really have a proper conversation. You’ll still need to change the contextWindow in the config file later. Unless you understand this issue and know how to fix it, we recommend not using this method until the official team fixes it. (Bug found in openclaw 2026.2.26)

1. Open the configuration wizard in openclaw onboard

When you reach the step to choose an AI model provider, select [Custom Provider].

2. Set the API BASE URL

If your Lobster is deployed overseas or you have a proxy/VPN enabled, use cloud; otherwise, cn and com are recommended.

3. Paste the API KEY

The sk--prefixed credential purchased from the 4All API website is called the API KEY, also known as the KEY or token. Paste it in and press Enter.

4. Choose the interface format

In 99% of cases, choose the first option, OpenAI format, because our site is an aggregated proxy API platform and most models support or are compatible with this format.

Anthropic-compatible is the format for the Claude series. If you are using a Gemini series model, you can choose Unknown and enter google-generative-ai

Unless you know exactly what you’re doing, just choose the first option.

5. Enter the available models

Enter the models available in the package you purchased. You can only enter one model here (this is a bit annoying; in [Option 5], Tencent Cloud’s simplified configuration likely uses the same command, which also only allows one model, so that’s a bit of a trap).

6. Customize the provider name

Usually, just enter 4allapi, as I do. It’s just an alias. After this is configured, when selecting a provider and model later, just look for 4allapi.

Model alias is an alternate name for the model; usually you can just press Enter to skip it.

Skip all subsequent steps.

7. Select the model to use

If this is your first configuration, openclaw will use the first model you just set as the default model, and you can start using it immediately without doing anything else.
If you have already configured a model from an official provider or another provider, you need to run openclaw onboard again. At that point, the provider 4allapi will appear at the bottom of the list. Then just select the model.

---

Option 3: Configure it through the web interface

1.1 Local installation and deployment

Usually, when the setup wizard completes for the first time, it will show a web address in the format http://ip:18789?token=xxxxxx

(If it’s not the first time and you forgot it, run the command openclaw dashboard to display the web address.)

As shown below, copy either of the two addresses marked in red and open it in your browser. The author deployed Lobster in WSL on Windows, so 127.0.0.1 and localhost work. If

1.2 Remote deployment such as on a cloud server

Because openclaw is designed rather strangely, for security reasons it does not allow the browser to directly access the console interface via IP + port. You need to use forwarding or an SSH tunnel. Refer to this tutorial provided by Tencent Cloud, or simply follow [the cloud server config-file method], which is the most convenient!

As shown above, Config is divided into from (visual form configuration) and raw (JSON editing configuration). As Lobster users, we definitely go straight to JSON configuration. Of course, you can also use the visual form; the result is the same:

2. Define the model providers and model list

Find the models node and add a custom model provider and models. Here I added three of the most popular Lobster-related models recently.

{
 "models": {
 "mode": "replace",
 "providers": {
 "4allapi": {
 "baseUrl": "https://api.4allapi.com/v1", // If the server is overseas or you have a proxy/VPN enabled locally, use api.4allapi.cloud
 "apiKey": "sk-xxxxx", // Enter the API KEY here
 "auth": "api-key",
 "api": "openai-completions", // For Claude-series models, use anthropic-messages; for Gemini-series models, use google-generative-ai. For most other models, use openai-completions by default
 "authHeader": true,
 "models": [
 {
 "id": "kimi-k2.5",
 "name": "kimi-k2.5",
 "api": "openai-completions",
 "reasoning": false,
 "input": [
 "text"
 ],
 "cost": {
 "input": 0,
 "output": 0,
 "cacheRead": 0,
 "cacheWrite": 0
 },
 "contextWindow": 256000,
 "maxTokens": 8192
 },
 {
 "id": "qwen3.5-plus",
 "name": "qwen3.5-plus",
 "api": "openai-completions",
 "reasoning": false,
 "input": [
 "text"
 ],
 "cost": {
 "input": 0,
 "output": 0,
 "cacheRead": 0,
 "cacheWrite": 0
 },
 "contextWindow": 256000,
 "maxTokens": 8192
 },
 {
 "id": "minimax-m2.5",
 "name": "minimax-m2.5",
 "api": "openai-completions",
 "reasoning": false,
 "input": [
 "text"
 ],
 "cost": {
 "input": 0,
 "output": 0,
 "cacheRead": 0,
 "cacheWrite": 0
 },
 "contextWindow": 256000,
 "maxTokens": 8192
 }
 ]
 }
 }
 }
}

3. Configure the default model

Find the agents node and set the default model.

{
 //前面其他配置........
 "agents": {
 "defaults": {
 "model": {
 "primary": "4allapi/kimi-k2.5" // Just change this line. Replace it with whichever of the models defined above you want to use
 },
 "models": {
 "4allapi/kimi-k2.5": {}
 },
 "workspace": "/root/.openclaw/workspace",
 "compaction": {
 "mode": "safeguard"
 },
 "maxConcurrent": 4,
 "subagents": {
 "maxConcurrent": 8
 }
 }
 }
 //后面其他配置.....
}

4. Verify that it took effect

Go back to the chat interface and have a casual conversation. If it responds, the configuration has taken effect. Then you can continue exploring openclaw using its various tutorials~

---

Option 4: Configure via the openclaw.json config file

This is the most direct method. If you have some technical background, or you’re a programmer, you can use this method directly.

1. Find the config file path

Linux/MacOs/Wsl subsystem:

echo $HOME/.openclaw

Windows CMD command:

echo %USERPROFILE%\.openclaw

Windows PowerShell command:

"$env:USERPROFILE\.openclaw"

2. Edit the config file

Here I use VS Code to open the .openclaw folder and config file inside the Windows WSL subsystem as an example. You can also use any editor that can format JSON, or go full vim to configure the JSON file. The configuration content is the same as in [Configure it through the web interface].

The biggest advantage of editing the config file directly, in my opinion, is that you can create multiple backups of openclaw.json under .openclaw by renaming files, and restore any previous configuration at any time!!!

---

Option 5: Configure via the cloud server config file

1) Tencent Cloud example: log in to the cloud server.

2) In the file manager, find the installation directory for Lobster/openclaw, locate openclaw.json, and edit it

Due to limited space, this section only demonstrates editing through Tencent Cloud’s web terminal. The process is similar on other cloud providers. If you have technical experience, using finalshell, xshell, or other SSH tools to modify the server remotely works just as well, so I won’t show that here.

---

Option 6: Tencent Cloud simplified openclaw configuration

1. Find the custom model form configuration

2. Configure the provider and model

After clicking [Add and Apply], if the configuration you just made appears in the app, then it worked.

This method is more beginner-friendly, but if you look closely you’ll notice it’s a bit clumsy and fairly restrictive. It only lets you configure one custom model at a time, and the next time you configure another one, it will overwrite the previous one. Also, it is no different from the [cloud server config file configuration] method in essence; both are just indirect ways of modifying openclaw.json.

---

Summary of large-model API configuration

If you think it through carefully, integrating 4All API’s large-model APIs into openclaw really only takes two steps.

1. Define models, meaning you need to customize the 4allapi provider and its model list

2. Set the model you want to use under the agents node (it must already have been defined in step 1)

3. Tip: You can actually just tell Lobster the URL, key, and model directly and let the AI configure it itself. Of course, this carries risk; unless you know exactly what you’re doing, manual configuration is still safer.

---

3. Advanced configuration

1. Configure multiple agents

Under the agents node, paste the following code at the same level as defaults (just follow the format below; you decide how many agents to configure and what to call them).

"list": [
 {
 "id": "main",
 "default": true,
 "name": "个人助手",
 "workspace": "~/.openclaw/workspace"
 },
 {
 "id": "code",
 "name": "编程助手",
 "identity": {
 "name": "kimi code",
 "theme": "Coding assistant",
 "emoji": "💻"
 },
 "model": "4allapi/kimi-for-coding",
 "workspace": "~/.openclaw/workspace-code"
 }
]

After saving, it can be used right away. You can also check the web interface to confirm whether it was added successfully. The result is as follows:

By the way, openclaw now has a Chinese interface, which makes things even easier.

---

2. Set up a separate agent for scheduled tasks

Here, we’ll demonstrate creating a new Job in the UI and assigning an Agent. Depending on the complexity and functionality of the scheduled task, you can assign different Agents, and the corresponding model pricing and usage will also differ, which can help save costs.

In addition to the UI method, you can also use the config file, or simply chat with Lobster and have it create a Job with the specified Agent for you. There are many ways to do this, so we won’t go into more detail.

---

4. Installation guide (if needed)

There are now installation tutorials for Lobster all over the internet, but their quality is indeed uneven. If needed, you can also refer to this document.

Linux/Wsl/Mac operating systems

curl -fsSL https://openclaw.ai/install.sh | bash

Windows operating system

Enter the following command in PowerShell or CMD (must be opened with administrator privileges):

iwr -useb https://openclaw.ai/install.ps1 | iex

If you know how to develop, you can complete the installation and configuration in VS Code or any IDE with a terminal. (Personally, I like doing it this way because it combines command entry and file inspection in one place.)

---

Screenshot walkthrough

1. Start installation

2. Select QuickStart and press Enter to continue installation

3. Choose an AI model provider.

Claude is the official top recommendation. However, since the providers here are all official platforms from major vendors, their pricing is generally expensive, or they require a proxy/VPN to access, which makes usage harder. So just choose Skip for now and configure the aggregated API platform 4All API directly later. If you already have an API from one of the listed platforms, you can choose it yourself.

4. Choose a model

By default, select All providers and press Enter.

If you already have an API from one of the listed platforms, choose the corresponding model family. For example, if you selected Moonshot AI in the previous step, you should choose their kimi-coding here.

Find the model. If you want to integrate 4All API or any model from a provider not included in Openclaw’s provider list, choose Enter model manually.

Then press Enter to proceed to the next step (configure it after installation is complete).

5. Choose Channel, Skills, and Hooks

The Channels are all overseas, so just skip them.

Skills and Hooks can also be skipped for now; you can configure them later.

6. Open the Web interface

After copying the link into your browser, you’ll be able to see the interface~

7. Select Open the Web UI and press Enter

Installation complete!!

---

4All API - One-stop AI large-model API aggregation platform
Official site: https://4allapi.com
API Base: https://api.4allapi.com