Why is my AI agent reusing cached API time responses instead of fetching fresh data?

The issue you’re experiencing is due to the conversation memory “remembering” the first API response and then reusing it in subsequent interactions. In other words, once the API response is stored in the chat memory, the AI will see that as part of its context and will not trigger the HTTP tool again because it “knows” the answer.

Here are a few approaches to solve the problem:

1. Isolate Factual API Calls from Chat Memory

If you need the API result (i.e., current time) to always be fresh, you should configure your workflow so that the API call’s response is not permanently stored in the chat memory. In practice, this means you can disconnect or avoid linking the HTTP tool’s output as part of the persistent memory. Some options for doing so are:

• Using a Separate Memory Channel: Instead of storing all outputs into the standard memory buffer, use a dedicated node or a conditional branch that bypasses storing data you expect to change continuously.
• Clear Specific Memory Entries: Utilize the Chat Memory Manager node to explicitly clear or override certain parts of the memory that shouldn’t persist. This way, whenever the conversation starts a new request for the current time, the previous result is not considered.

2. Reset or Truncate the Memory for Ephemeral Data

You might consider performing a memory reset at the start of your workflow for those attributes that should not be reused. For example, if you are using the Window Buffer Memory node, check if you can limit its capacity or configure it such that temporary API responses (like the current time) are excluded. This ensures that every new request starts with a fresh state.

3. Force a Fresh API Call via Tool Configuration

If the intent is to always fetch the latest time regardless of memory, then adjust the prompt or agent parameters so that it explicitly ignores stored answers for this type of query. Although you mentioned trying to force the reques without success, check:

• If your prompt includes instructions to bypass cached data, verify that the agent’s logic is set up to rerun the tool even if similar data exists in memory.
• Ensure the HTTP tool call is directly linked to the AI agent as an “ai_tool” action rather than being indirectly provided from prior memory data.

Summary

Your AI agent is not invoking the API a second time because its chat memory is holding on to the first response. By separating volatile data (like a current time request) from the long-term chat memory, or by clearing/overriding this memory for such cases, you can force the agent to reissue the HTTP tool call for an updated response.

If you need more advanced memory management, consider using the Chat Memory Manager node, which provides greater control over what gets stored and when it should be reset.

Let me know if you need more detail on any of these options!