Skip to main content
AgentScope provides unified async abstractions for various AI models across different providers:

Chat Models

Core text generation with reasoning, streaming, and tools API support.

TTS Models

Convert text to speech with realtime and non-realtime options.

Realtime Models

Bidirectional WebSocket streaming for low-latency voice agents.

Embedding Models

Generate vector representations for retrieval and similarity search.

​
Chat Model

Chat models are the core of the agent, enabling it to generate streaming/non-streaming responses, perform reasoning, and call tools.
The streaming mode in AgentScope chat models is accumulative — each yielded response contains all content generated so far, not just the latest delta. This design simplifies consumption since you always have the complete current state without tracking deltas.
To support multi-agent conversations in a chatbot format, AgentScope designs a formatter layer that
  • converts AgentScope’s Msg objects into the expected input format for each LLM API, and
  • adopts multi-agent conversation context into the two-role chatbot format by prefixing messages with agent names and wrapping them in <history> tags.
Such formatters are distinguished by the suffix ChatFormatter (e.g., DashScopeChatFormatter) and MultiAgentFormatter (e.g., DashScopeMultiAgentFormatter) — the former is for two-party conversations (user + assistant), while the latter is for multi-agent conversations.
For detailed usage examples and a full provider reference table mapping each model class to its corresponding formatter, see Models.

​
TTS Model

TTS (Text-to-Speech) models convert text into audio. AgentScope supports both non-realtime and realtime TTS models:

​
Realtime Model

Realtime models provide bidirectional, persistent communication over WebSocket, designed primarily for voice agent scenarios where the user speaks and the model responds with speech in real-time.

​
Embedding Model

Embedding models generate vector representations for text, images, and other data types. These embeddings are used for retrieval, similarity search, and as input features for downstream tasks.