RegisterAgent

Possible values: [OpenAIChat, OpenAIResponses]

Default value: OpenAIChat

The vendor type of the LLM interface.

• OpenAIChat: OpenAI's Chat Completions interface type.
• OpenAIResponses: OpenAI's Responses API interface type.

The endpoint that receives the request (can be your own service or any LLM service provider's service). If Vendor is OpenAIChat, it must be compatible with OpenAI Chat Completions API. For example: https://api.openai.com/v1/chat/completions

If Vendor is OpenAIResponses, it must be compatible with OpenAI Responses API. For example: https://ark.cn-beijing.volces.com/api/v3/responses

📌 Important Note

If ApiKey is set to "zego_test", you must use one of the following Url addresses:

• MiniMax：https://api.minimax.chat/v1/text/chatcompletion_v2
• Volcano Engine (Doubao): https://ark.cn-beijing.volces.com/api/v3/chat/completions
• Aliyun Bailei (Tongyi Qianwen): https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions
• Stepfun: https://api.stepfun.com/v1/chat/completions

The parameter used for authentication by the LLM service provider. It is empty by default, but must be provided in production environments.

📌 Important Note

During the test period (within 2 weeks after the AI Agent service is enabled), you can set this parameter value to "zego_test" to use this service.

The LLM model. Different LLM service providers support different models, please refer to their official documentation to select the appropriate model.

📌 Important Note

If ApiKey is set to "zego_test", you must use one of the following models:

• MiniMax:
  • MiniMax-Text-01
• Volcano Engine (Doubao):
  • doubao-1-5-pro-32k-250115
  • doubao-1-5-lite-32k-250115
• Aliyun Bailei (Tongyi Qianwen):
  • qwen-plus
• Stepfun:
  • step-2-16k

The system prompt of the AI agent. It is the predefined information that is added at the beginning when calling the LLM, used to control the output of the LLM. It can be role settings, prompts, and answer examples.

Possible values: >= 0 and <= 2

Default value: 0.7

The higher the value, the more random the output; the lower the value, the more concentrated and determined the output.

Possible values: >= 0 and <= 1

Default value: 0.9

The sampling method. The smaller the value, the stronger the determinism; the larger the value, the stronger the randomness.

Other parameters supported by the LLM service provider, such as the maximum token limit. Different LLM providers support different parameters, please refer to their official documentation and fill in as needed.

Default value: false

If this value is true, the AI Agent server will include the AI agent information(agent_info) in the request parameters when requesting the LLM service. The example of the AI agent information is as follows: Using Custom LLM. You can use this parameter to execute additional business logic in your custom LLM service.

The structure of agent_info is as follows:

• room_id: RTC room ID
• user_id: User ID
• agent_instance_id: AI agent instance ID

📌 Important Note: Only effective when Vendor is "OpenAIChat".

Possible values: [Aliyun, ByteDance, ByteDanceV3, ByteDanceFlowing, MiniMax, CosyVoice]

The TTS service provider. Please refer to Configuring TTS > TTS Parameters for details.

Possible values: <= 4 characters

Can be used to set the termination text for TTS. If the input to TTS (usually the content returned by LLM or the Text parameter of the SendAgentInstanceTTS API) contains the TerminatorText string, then the content from the TerminatorText string (inclusive) onward will no longer be synthesized in this TTS round.

📌 Important Note

• For bidirectional streaming, only one character can be set.

• Typically, the LLM is guided via prompts in LLM > SystemPrompt to specify which parts of the content should be enclosed with special punctuation.

• This parameter cannot be updated when updating an agent instance.

The specified strings in the content input to TTS (usually the content returned by LLM or the Text parameter of the SendAgentInstanceTTS API) will not be involved in speech synthesis. Each string in the array indicates a string to be filtered out, and each string can have up to 2 characters.

Possible values: [Tencent, AliyunParaformer, AliyunGummy, Microsoft]

Default value: Tencent

ASR provider. Please refer to Configuring ASR > ASR Parameters for details.

Vendor parameters, please refer to Configuring ASR > Params Parameters for details.

Possible values: >= 200 and <= 2000

Default value: 500

Set the number of seconds after which two sentences are no longer considered as one. Unit is ms, range [200, 2000], default is 500. Please refer to Speech Segmentation Control for details.

📌 Important Note

When Vendor is "Tencent" (default), the maximum value is 1500.

Possible values: >= 200 and <= 2000

Set the number of seconds within which two sentences are considered as one, i.e., ASR multi-sentence concatenation. Unit is ms, range [200, 2000]. Only when this value is greater than VADSilenceSegmentation, ASR multi-sentence concatenation will be enabled. Please refer to Speech Segmentation Control for details.

Possible values: [0, 1, 2, 3]

Default value: 0

VAD sensitivity, 0: medium sensitivity, default value; 1: low sensitivity; 2: high sensitivity; 3: custom mode, needs to be used in combination with VADMinSpeechDur and VADEnergyThreshold. Please refer to Voice Interruption Sensitivity Adjustment for details.

Possible values: >= 0 and <= 1000

VAD minimum speech duration, unit ms, the larger the value, the less likely it is to be detected, but it may cause some short speech to be missed, the value range is [0, 1000].

Possible values: >= 0 and <= 1

VAD energy threshold, the smaller the value, the higher the sensitivity, the value range is [0, 1].

This parameter has been deprecated. Please set it through the Params vendor parameters.