System instrcution in langgraph and Restful API

[AnemonePetal]

Hi everyone. I know it sounds like a trivial question. But it make me very confused.

According to my observation (using langsmith), langgraph tends to handle system instruction like:

# {
#   "messages": [
#     {"role": "system", "content": "You are a helpful assistant..."},
#     {"role": "user", "content": "Previous user message"},
#     {"role": "assistant", "content": "Previous assistant response"},
#     {"role": "user", "content": "Current user message"}
#   ]
# }

However, if we check the implementation of gemini cli. It actually use systemInstruction field in the method generateContent. If we use json to represent each field, I could give an example:

  model: 'gemini-2.0',
  contents: [
    // environment prompt
    {
      role: 'user',
      parts: [{ text: "This is the Gemini CLI. We are setting up..." }]
    },
    {
      role: 'model',
      parts: [{ text: "Got it. Thanks for the context!" }]
    },
    // Previous conversation turns (if any)...
    // Real user message
    {
      role: 'user',
      parts: [{ text: "Your question" }]
    }
  ],
  config: {
    systemInstruction: { 
      text: "You are an interactive CLI agent specializing in software engineering tasks..." 
    },
    thinkingConfig: { includeThoughts: true },
    tools: [{ functionDeclarations: [...] }]
  }

You could see that gemini cli would also provide environment prompt at the beginning of the chat. However, role definition would be put into their native systemInstruction field.

I want to know whether langgraph support this way to provide systemInstruction? Do we have any benefits to do that? How does google handle this field systemInstruction?