Am I the only one who finds the agent 'memory' feature totally unreliable?

Hey everyone! Has anyone else been wrestling with AgentGPT's memory system? I've been trying to build a simple research assistant agent that can summarize articles over multiple sessions, and the memory feature just doesn't seem to persist things reliably for me. 😕

I'm setting up my agent with what looks like the correct configuration:

```yaml
agent_name: "ResearchSummarizer"
goal:
- "Read and summarize key points from provided tech articles"
- "Maintain a list of summarized articles to avoid repeats"
memory: true
model: "gpt-4"
```

The agent *claims* it's storing facts in memory, like "Already summarized article X on quantum computing." But when I start a new session with the same agent name and goal, it often acts like it's seeing a topic for the first time. It feels like the memory is either being cleared unexpectedly or the retrieval is failing silently.

I've noticed a few patterns:
* Short-term memory within a single session *mostly* works.
* The moment you restart the agent (even with identical config), it's a coin toss.
* There's no clear way to "peek" at what's actually stored, which makes debugging guesswork.

This makes multi-step workflows or context-heavy agents really brittle. I tried to use it for a code review agent that remembered common style issues in our project, but I couldn't rely on it.

Are we missing something? Is there a specific way this needs to be set up, or are we all just waiting for a more stable implementation? I'd love to hear your experiences and any workarounds you've found.