[Bug]: always exceeds context limits, and then crash the models

[francogp]

Bug Description

I tested 32k, 64k, 128k, and 256k models. Hermes always exceeds the maximum and stops responding until I start a new conversation. It never compact when reaches half context.

Steps to Reproduce

just use it with gemma 4 with context limits at 32k, 64k, 128k, and 256k.

Expected Behavior

to compact the context so it can continue to work

Actual Behavior

it exceeds the limits and STOP responding (the model)

Affected Component

Agent Core (conversation loop, context compression, memory)

Messaging Platform (if gateway-related)

N/A (CLI only)

Debug Report

Report       https://paste.rs/Of0VQ
  agent.log    https://paste.rs/8OMbC
  gateway.log  https://paste.rs/PQ2oq
  gui.log      https://paste.rs/jgBKx

Operating System

Ubuintu 26.04

Python Version

Python 3.13.5

Hermes Version

Hermes Agent v0.18.0 (2026.7.1) · upstream 63354ed

Additional Logs / Traceback (optional)

Root Cause Analysis (optional)

No response

Proposed Fix (optional)

No response

Are you willing to submit a PR for this?

• I'd like to fix this myself and submit a PR

[githubnoobie916]

This is also happening to me. Its making hermes agent unusable.

[githubnoobie916]

I am using codex 5.6 on high with thinking enabled

[Widiskel]

same, im using codex 5.5