Bug Description

When the context compression runs with summary_model_override=None (the default), and the Groq API returns a 413 TPM error, the code does NOT retry on the main model (Qwen3 local). It just waits for cooldown and returns None.

This happens because in context_compressor.py, line ~939:

if (self.summary_model 
    and self.summary_model != self.model):

When summary_model_override=None is set, this becomes an empty string which is falsy. The condition short-circuits: never enters the retry block for main model.

So when Groq hits 413 (TPM exhausted), there's no fallback to Qwen3 local.

Root Cause

The code assumes a truthy summary_model name, but None becomes empty string which is falsy. The condition should also check if summary_model_override was never set (is not None) and retry on main model in that case.

Evidence (from logs)

ContextCompressor summary failed with HTTP 413: Requested tokens exceed the maximum tokens per minute for this model.
ContextCompressor cooldown: 60s before next retry attempt
ContextCompressor no more retries, returning None (compressed 0 messages)

No Qwen3 fallback appears between Groq retries — confirms condition is False when summary_model="".

Proposed Fix

The fix should check self.summary_model_override is not None separately:

if self.summary_model or True:  # always enter retry block
    if (not self.summary_model and self.model) or (self.summary_model != self.model):
        # Retry on main model when no override set, OR different models configured

Impact

• Groq free tier users with summary_model_override=None get zero compression once TPM is exhausted
• Context keeps growing until session restarts