Requirements#
GenAI Execution
|
status: valid
security: NO
safety: QM
|
||||
The platform shall support the execution of Generative AI models (e.g., LLMs) on embedded automotive hardware. |
|||||
GenAI Interaction Layer
|
status: valid
security: YES
safety: QM
|
||||
The platform shall provide a Prompting Interface for GenAI-based agents, enabling structured prompts, streaming output, and context-aware user interaction. |
|||||
GenAI Action Validation
|
status: valid
security: YES
safety: ASIL_B
|
||||
The platform shall validate all LLM-generated actions via domain-specific policies before they are executed. |
|||||
Structured Vehicle Interface
|
status: valid
security: YES
safety: ASIL_B
|
||||
The platform shall provide structured APIs (e.g., via MCP) to access vehicle state and invoke safe vehicle commands. |
|||||
GenAI Context Management
|
status: valid
security: YES
safety: QM
|
||||
The platform shall provide a Context API that manages short-term context (e.g. current goal, location, dialogue state) and long-term context (e.g. driver preferences, history) for GenAI-based agents, decoupled from low-level system interfaces such as file system or persistent storage. |
|||||
Scoped Long-Term Context Updates
|
status: valid
security: YES
safety: QM
|
||||
The platform shall restrict write operations to long-term context to explicitly scoped and validated updates. |
|||||
Prompt Input Validation
|
status: valid
security: YES
safety: QM
|
||||
The platform shall validate and rate-limit prompt inputs to the GenAI runtime to protect against malformed sequences and prompt injection. |
|||||
Concurrent Model Execution
|
status: valid
security: NO
safety: QM
|
||||
The platform shall support the concurrent execution of multiple GenAI models. |
|||||
GenAI Runtime Neutrality
|
status: valid
security: NO
safety: QM
|
||||
The platform shall integrate established GenAI model runtimes via a runtime-agnostic interface and shall not provide a proprietary GenAI runtime implementation. |
|||||
MCP Access Control
|
status: valid
security: YES
safety: QM
|
||||
The platform shall authenticate and authorize all communication between MCP Client and MCP Server before vehicle state is exposed or vehicle commands are invoked. |
|||||