This page was machine-translated and may differ from the original. View original

Automatic completion of CAPL test generation, execution, and error correction within minutes using natural language prompts
A new feature has been added to CANoe that automates the entire vehicle network development and testing process with a single natural language input. Vector Korea announced the launch of a new version integrating AI agents and MCP (Model Context Protocol), enabling test tasks that previously took hours to days to be completed within minutes.
Vector Korea (Branch Director Jang Ji-hwan) officially launched 'CANoe AI Package (CANoe AI Package)' with AI agent and MCP functionality in 'CANoe 20 SP2,' a development and testing environment for vehicle networks and embedded systems, on the 27th.
The company explained that the package is available for free download from Vector's official website and is also compatible with users' own large language model (LLM) environments.
■ Test Automation Based on Natural Language Prompts
The core of this update is a structure where the AI agent automatically performs the entire process from requirements analysis to CAPL test code generation, execution, error correction, and re-testing with a single natural language prompt.
After task completion, users can verify results and grant final approval on the synchronized CANoe screen.
Vector Korea stated that users can directly set the autonomy level of the AI agent, and all task steps can be monitored in real-time within the CANoe development environment.
An open AI layer is also provided to freely connect existing LLMs such as GitHub Copilot and Claude.
The company explained that this enables domain-specific customization and extensions utilizing agent, skill, and MCP tool components.
■ Hallucination Suppression through Vector-RAG Technology
Vector Korea announced that by integrating 'Vector-RAG (Retrieval-Augmented Generation)' technology, its document and technical reference search system, into the package, it has suppressed hallucination phenomena that can occur in general-purpose language models.
Through this, code and answers can be generated based on verified expert knowledge, and functions such as △configuration inquiry and customization △simulation control △test case generation △CAPL, C#, and Python code generation and optimization can be performed with a single natural language command.
Vector is headquartered in Stuttgart, Germany, and has more than 4,500 employees worldwide.
2024 revenue reached €1.01 billion.
본 기사에 대한 정정·반론·추후보도 청구는 보도 청구 안내를, 그간 게재된 보도문은 정정·반론보도 모아보기를 참고해 주세요.















