hermes - 💡(How to fix) Fix [Bug]: conversational amnesia when running gemma4 models

Bug Description

I am experimenting on a old gaming rig.

It's running a old RTX 2070 Super with 8GB VRAM and 16GB RAM

Environment is a update to date windows 10 WSL2, Ubuntu

So I was playing around with ollama running gemma4:latest hooking as the default model provider.

After playing around on hermes agent, I noticed it was not remembering the conversation we were having at all.

At first I thought it might be a memory issue so I down graded to gemma4:e2b which was 7.2 GB.

However same issue persisted.

The test I used was the following:

1. ollama serve
2. ollama run gemma4:e2b
3. hermes model -> select ollama -> select gemma4:e2b
4. hermes
5. typed the following messages:

"hello" "I am going to run a memory test on you" "please tell me what was the last message that I typed to you"

for both gemma4:e2b and gemma4:latest the response was some thing like:

I do not have access to the specific history of our conversation outside of the current interaction context when you ask a question. Please provide the previous message you are referring to, or clarify what you are asking about.

which was strange, as prompting the model directly in ollama does not seem to have the same problem.

further more I switched the model to qwen3.5:latest and the test works correctly, both in hermes agent and ollama chat.

This suggests there is some interface / configuration issue with ollama and hermes agent in regards to gemma4

Steps to Reproduce

Hardware:

RTX 2070 Super with 8GB VRAM and 16GB RAM

Environment:

Update to date windows 10 WSL2, running Ubuntu

Steps:

1. install ollama default options, install hermes agent default options.
2. ollama serve
3. ollama run gemma4:e2b
4. hermes model -> select ollama -> select gemma4:e2b
5. hermes
6. typed the following messages:

"hello" "I am going to run a memory test on you" "please tell me what was the last message that I typed to you"

for both gemma4:e2b and gemma4:latest the response was some thing like:

"I do not have access to the specific history of our conversation outside of the current interaction context when you ask a question. Please provide the previous message you are referring to, or clarify what you are asking about."

which was strange, as prompting the model directly in ollama does not seem to have the same problem.

further more I switched the model to qwen3.5:latest and the test works correctly, both in hermes agent and ollama chat.

Expected Behavior

the agent will return a message stating the last message was "I am going to run a memory test on you"

Actual Behavior

for both gemma4:e2b and gemma4:latest the response was some thing like:

"I do not have access to the specific history of our conversation outside of the current interaction context when you ask a question. Please provide the previous message you are referring to, or clarify what you are asking about."

which was strange, as prompting the model directly in ollama does not seem to have the same problem.

further more I switched the model to qwen3.5:latest and the test works correctly, both in hermes agent and ollama chat.

Affected Component

Agent Core (conversation loop, context compression, memory)

Messaging Platform (if gateway-related)

N/A (CLI only)

Debug Report

Report     https://dpaste.com/4E3K3MSPV
  agent.log  https://dpaste.com/97NMK75AT

Operating System

Windows 10, WSL2, Ubuntu 20.04.4 LTS

Python Version

Python 3.8.10

Hermes Version

Hermes Agent v0.15.1 (2026.5.29)

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