Real-time Voice Interaction
Supports a conversation rhythm of 'understanding while speaking, interruptible, and resumable' with a real-time streaming voice processing framework, making interactions closer to human conversations.
Utilizes real-time streaming processing and latency control to shorten the wait time from speaking to response, making conversations smoother and more continuous.
Can detect user interruptions in real-time and stop the current response, naturally continuing after preserving context, reducing the feeling of having to wait for the AI to finish speaking.
Automatically recognizes and switches between multiple languages, maintaining stable recognition even in noisy environments, allowing smooth interaction with users from different countries.
Tool Calling
Transforms each step of the service process into executable tools, allowing conversations to not only stop at responses but also complete queries, updates, triggers, and status feedback within tasks.
Packages enterprise data and processes into callable tools, defining input and output with structured parameters, enabling AI to execute queries, updates, and process triggers more accurately and consistently.
AI can determine when to call tools during conversations, proceeding to the next step after obtaining results, advancing tasks from single responses to actual completion.
Retains input, return, and error records for each tool call, supporting retries, supplementary inquiries, or manual takeover, making tasks more stable and interchangeable in real scenarios, maintaining context consistency, and ensuring uninterrupted interaction.
Multimodal UX
Not only handling voice, but also coordinating key points and next steps with the interface, allowing conversations, visuals, and operations to collaborate, making task advancement more natural and smoother.
Uses visual key screens to capture voice conversation results, helping users quickly understand AI response highlights and next step judgments, reducing the comprehension cost of relying solely on voice interaction.
Converges input with structured components, and generates appropriate interfaces based on context with AI, allowing undefined interaction needs to be flexibly accommodated.
Naturally switches between voice, text, maps, images, and videos, maintaining context consistency and ensuring uninterrupted interaction.
Multi-model Orchestration
Decomposes executable steps with a task orchestration mechanism, collaborating and scheduling among multiple models according to needs, balancing quality, cost, and latency, resulting in more stable and controllable outcomes.
Decomposes, sequences, and assigns tasks to different specialized capabilities for collaboration, then consolidates them into a consistent deliverable result.
Sets quality checkpoints for key steps, retaining decision and execution records for traceability and subsequent optimization.
Dynamically selects models based on task steps and needs, balancing quality, cost, and latency, avoiding reliance on a single model for all scenarios.
