How do I work with the BufferMemory?

[JosefKousal]

So I have this code:

const chain = new ConversationChain({
            ChatPromptTemplate.fromPromptMessages([
              SystemMessagePromptTemplate.fromTemplate(
                  `Some template ${arg1}`
                ),
              SystemMessagePromptTemplate.fromTemplate(
                  `Conversation history: {history}`
              ),
              HumanMessagePromptTemplate.fromTemplate("{arg2}"),
          ]),
            memory: new BufferMemory({
                memoryKey: "history",
                inputKey: "input",
            }),
        })
        
const customerResponse = await chain.call({ 
    "arg1": "arg1",
    "arg2": "arg2",
})
const memory = chain.memory
const otherChain = new ConversationChain({
    llm: this.langchainOpenAI,
    memory: memory,
    prompt: ChatPromptTemplate.fromPromptMessages([
        SystemMessagePromptTemplate.fromTemplate(
            `Some text`
        )
    ])
})
const otherResponse = await otherChain.call({})

Given that this is BufferMemory, I'd expect it to collect all the messages and append them one after the other - including the System- and Humanmessages passed into the first chain.call. Instead, I only get the previous answer from the AI:
ChatMessageHistory:

messages: [
    HumanChatMessage { text: undefined, name: undefined },
    AIChatMessage {
      text: 'Some response',
      name: undefined
    }
  ]

What's the catch? What am I doing wrong? I tried an LLMChain and a ConversationChain, both to no avail.

I also tried adding

SystemMessagePromptTemplate.fromTemplate(
            `Conversation history: {history}`
        ),

to the other prompt, which yielded the same results.

The context length should not be an issue, it's approx. 3000 tokens being fed into gpt-3.5-turbo (both requests and responses in total).