How to get Agent node's token usage in the next node?

[maxleton-cyber]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
• [FOR CHINESE USERS] 请务必使用英文提交 Issue，否则会被关闭。谢谢！:)
• Please do not modify this template :) and fill in all the required fields.

1. Is this request related to a challenge you're experiencing? Tell me about your story.

I'm using Dify 1.16.1's new sandboxed "Agent" node (not the classic Agent node) in a workflow. The Agent node only exposes two output variables — text and json — with no native token/usage output variable. However, as shown in the trace panel, the Agent node clearly tracks and displays its token consumption (e.g. "252.421K tokens") internally.

I need to pass this token usage value to the next node (a code node) so I can calculate credit consumption based on it. Right now I have no way to reference the Agent node's actual token usage from downstream nodes — I can only get it by asking the LLM to report it inside its own JSON output, which is unreliable (the model sometimes miscounts or omits it).

Is there a way to expose the Agent node's real token/usage stats as an output variable, similar to how LLM nodes expose usage?

2. Additional context or comments

Screenshot attached showing the trace panel where the Agent node's token usage (252.421K tokens) is visible in the UI, but this value doesn't appear to be accessible as a node output variable in the workflow.

[dosubot[bot]]

@maxleton-cyber the usage data actually is being written to the Agent node's output dict on the backend — it's just not surfaced in the UI's variable picker yet. In message_transformer.py, when the Agent node completes, it yields a StreamCompletedEvent whose outputs include:

outputs={
    "text": text,
    "usage": jsonable_encoder(llm_usage),   # ← prompt_tokens, completion_tokens, total_tokens, prices, etc.
    "files": ...,
    "json": json_output,
    **variables,
}

So the underlying token usage data (prompt_tokens, completion_tokens, total_tokens, pricing) is already being computed and stored [1]. The gap is that the frontend doesn't declare usage as a selectable output variable for the Agent node, so you can't reference it from a downstream Code node through the normal variable picker.

This is exactly what your #40498 issue tracks [2].

In the meantime, since the data isn't exposed in the UI:

1. If you're self-hosting, you could patch the frontend to add usage to the Agent node's declared output variables so it appears in the picker.
2. Otherwise, keep an eye on New Agent node output does not expose LLM token usage #40498 — adding weight there (upvotes/comments) would help prioritize the fix.

To reply, just mention @dosu.

---

Share context across your team and agents. Try Dosu.