Hi everyone, I'm using

openhands-sdk

and am running into infinite loop of condensation as follows:

[01/02/26 15:17:16] WARNING  LLM raised context window exceeded error, triggering condensation                                                                                                  agent.py:200
Condensation Request ───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────

Conversation Condensation Requested
A condensation of the conversation history has been requested to manage context window usage.

Has anyone run into this issue and resolved it before? For context, I'm using the

LLMSummarizingCondenser

condenser = LLMSummarizingCondenser(
        llm=llm.model_copy(update={"usage_id": "condenser"}), max_size=200000, keep_first=2
    )

@wide-doctor-37830 is working on this I think!

Hey Seungone, there's a few possible causes for this (including some very real bugs I'm working on squashing). In the meantime one thing I see that'll trip up the system is that

max_size

references the number of events before condensation, not the number of tokens. 200k is huge -- I'd recommend setting that value to closer to 200.

@wide-doctor-37830 Oh I thought this was the token count, thanks for sharing!!