Docs

ArkClaw Enterprise

Convenience features

Adjusting/configuring inference models

ArkClaw supports subscriptions to either the Agent Plan or Coding Plan, as well as models available in the ModelArk Foundation models. It also allows custom integration with large model services provided by third-party vendors, such as the official DeepSeek service, and configuring fallback models for the primary agent. This enables ArkClaw to fully adapt to diverse personalized inference scenarios. You can flexibly select and configure the target model based on your subscription permissions.

tip

This topic may contain information about third-party products, which is provided for reference only. The Volcano Engine platform makes no representations or warranties of any kind, express or implied, regarding the performance, reliability, or availability of any third-party product, or any consequences arising from its use.

If you have created an agent team for ArkClaw, in addition to configuring an inference model for the primary agent, you can also configure independent inference models for other agents in the team. Learn more.

Adjust inference model

Switch between model services that your current account has subscribed to or is authorized to access for different business scenarios.

tip

The following operation applies to changing the primary model only. To modify fallback models, see Configure inference models.

Switch to the target ArkClaw page.

In the ArkClaw chat box, click the current model name.

Select the source channel for the target model (for example, Agent Plan and "Foundation models"), and click the target model in the corresponding model list to switch to the model.

tip

The model source channel is associated with account subscription permissions. Only subscribed model service channels are displayed.

To subscribe to or authorize access to other models if the current model does not meet your needs, refer to "Configure the model" in the next step.

(Optional) If you do not have permission to access the new model using the original API Key after switching to a model from Coding Plan or Foundation models, a message appears above the dialog box: "API Key expired. Please reconfigure it and proceed". Click "Configure now". In the Configure agent model dialog box, update the API Key or reconfigure other inference models.

Configure inference models

By default, the platform provides a set of general-purpose inference models (such as the entitled model included with trial token quota) used for agent capability configuration and calling directly. You can also integrate new inference model services, for example, by subscribing to the Agent Plan package, to provide ArkClaw with stable large model inference capabilities.

Limits

The range of model sources configurable in ArkClaw varies based on your subscribed large model subscription package (Agent Plan and Coding Plan).

Subscription status	Supported model configuration source
Not subscribed to any package	Agent Plan, Foundation models, Custom model configuration, and Trial token quota
Subscribed to the Agent Plan
Subscribed to the Coding Plan	Coding Plan, Foundation models, Custom model configuration, and Trial token quota
Subscribed to both packages	Agent Plan, Coding Plan, Foundation models, Custom model configuration, and Trial token quota

Prerequisites

If you need to integrate a large model service provided by a third-party vendor, ensure that you have access to the target model service and have prepared the model name, Base URL, and API Key for authentication in advance.

This topic uses DeepSeek, GLM, Kimi, and Qwen as examples to describe how to obtain the relevant information:

warning

Protocol compatibility limits: ArkClaw currently only supports integrating API Base URLs that are compatible with the OpenAI standard protocol.

Service and billing declaration: You need to prepare and maintain the integrated third-party large model services yourself. ArkClaw only provides a unified call channel. It does not provide the API Keys or Token quotas for third-party models.

Third-party model	Configuration parameters and acquisition methods
DeepSeek	Model name: Visit the official DeepSeek website to obtain it. For details, see Model list. Access address: https://api.deepseek.com API Key: Refer to DeepSeek API Key guide or log in to the DeepSeek console to obtain it.
GLM (Zhipu AI)	Model name: Visit the official Zhipu website to obtain it. For more information, see Model list. Access address: https://open.bigmodel.cn/api/paas/v4/ API Key: Refer to Getting started or log in to the GLM console to obtain it.
Kimi (Moonshot AI)	Model name: Visit the official Moonshot website to obtain it. For details, see Model list. Access address: https://api.moonshot.cn/v1 API Key: Refer to Obtaining an API Key or log in to the Kimi console to obtain it.
Qwen (Tongyi Qianwen)	Model name: Visit the official Alibaba Cloud website to obtain it. For details, see Model list. Access address: https://dashscope.aliyuncs.com/compatible-mode/v1 API Key: Refer to How to Get API Key or log in to the Qwen console to obtain it.

Procedure

Switch to the target ArkClaw page.

In the ArkClaw chat box, click the current model name and select the "Settings" button.

Configure the primary model or fallback models as needed.

warning

The fallback model configuration feature is currently available only for Agent Plan and Coding Plan packages.

Fallback models can currently be configured only for the primary agent. Other agents automatically inherit the primary agent configuration.

Configure primary model

Configure fallback models

The primary model is the model that an agent calls first during inference. The primary model takes priority for inference in every request. Only one primary model can be configured under the same configuration.

In the "Configure agent model" pop-up window, select the target model and refer to the table below to configure the relevant parameters. ArkClaw allows you to select a model from the following five channels: "Agent Plan", "Coding Plan", "Trial token quota", "Foundation models", and "Custom model configuration".

tip

The Agent Plan is a brand-new upgraded subscription package built on the basis of the Coding Plan. It is recommended as the preferred option.

Model source	Operation and description
Agent Plan	Subscribe to the Agent Plan package to provide ArkClaw with stable large model inference capability. (Optional) Click the "Go to activate" button and select the Agent Plan package you want to subscribe to. For the models supported by Agent Plan and model selection recommendations, see Plan benefits. tip Configuration is required only if you have not subscribed to the Agent Plan package. Click the "Agent Plan" card, and select the target model and its corresponding API Key from the drop-down list. Click "Confirm" to complete the operation.
Coding Plan	Adjust the subscribed Coding Plan package to provide ArkClaw with stable large model inference capability. tip If you are not subscribed to the Coding Plan, you will not be able to configure models supported by the Coding Plan. Click the "Coding Plan" card, and select the target model and its corresponding API Key from the drop-down list. For the models supported by Coding Plan and model selection recommendations, refer to Supported models. Click "Confirm" to complete the operation.
Trial token quota	Use the trial token quota to provide underlying inference services for ArkClaw. tip After you purchase ArkClaw, the platform automatically allocates a free trial token quota of 50M each month. The Doubao-Seed-2.0-lite model from the ModelArk Foundation models is configured by default for your trial use. Quota consumption and switch rules If you have purchased the large model subscription package (Agent Plan), the platform will deduct quota from the package first. You can also manually switch to the Doubao-Seed-2.0-lite model as needed to use the free trial quota. If you do not have the subscription package, the platform will consume the free trial quota by default. Once your trial token quota runs out, you may upgrade to Agent Plan or modify the model configuration to obtain more exclusive quota and higher configuration. Click the "Trial token quota" card. Confirm that there is still trial token quota available, and click the "Confirm" button to complete the operation.
Foundation models	Use a large model from the ModelArk Foundation models as the inference large model for ArkClaw. warning Using models from ModelArk "Foundation models" will not consume your Coding Plan quota. You will be charged based on token usage. Click the "Foundation models" card, and select the target model and its corresponding API Key from the drop-down list. Click "Confirm" to complete the procedure.
Custom model configuration	You can customize third-party large models to be integrated to provide underlying inference services for ArkClaw. In this section, "DeepSeek-V3.2" is used as an example for integration. Click the "Custom model configuration" card. Enter the model name, access address, and API Key for authentication. tip Ensure that the model name, access address, and API Key are strictly matched and not mixed (e.g., using a DeepSeek access address with a Zhipu AI API Key). Otherwise, authentication failure or call errors occur. Model name: Enter the target large model name, such as "DeepSeek-V3.2". For details, refer to Prerequisites. Access address: Enter the API interface call address (Base URL) of the target model service, such as "https://api.deepseek.com". For details, refer to Prerequisites. API Key: Enter the authentication key (API Key) used to call and access the target model service, such as "sk-j08hbgg5h3****". For details, refer to Prerequisites. warning The API Key will be used as an important credential for configuring ArkClaw's underlying model call service, granting full access permission to the model. Keep it safe and prevent leakage. If the API Key is leaked, it may cause unauthorized use of your model call quota or other security risks. Click "Confirm" to complete the procedure.

Last updated: 2026.08.26 18:19:28