[Feature Request] Persistent memory or session-level context management in Amazon Q CLI to reduce token usage

**Summary**

As a heavy user of the Amazon Q CLI under the Developer Pro subscription, I’ve found a significant limitation that directly impacts usability and efficiency: the lack of persistent memory or session-aware context in the CLI leads to extremely high token consumption — especially for workflows requiring continuity or logging.

**Problem**

Amazon Q CLI currently operates statelessly — each interaction starts from scratch, requiring the full system prompt and relevant context to be re-supplied with every request. This creates a major problem in real-world development workflows:

- I use the CLI to automate and streamline my coding tasks.
- I also ask it to log actions in a structured format (daily summaries, ongoing logs, etc.).
- Since there is no memory, I must resend full context or detailed instructions on every invocation.

This design results in **rapid exhaustion of my monthly quota**, often within days, despite being on the Pro plan. The problem isn’t the frequency of use, but the **inefficiency of repeated context transmission**.

**Feature Request**

Introduce one of the following improvements:

1. **Lightweight session memory**: Maintain context across CLI commands during a session, similar to conversational memory in ChatGPT or Claude.
2. **User-defined memory toggle**: Allow developers to optionally persist a system prompt or initial context (e.g., as a local config file or cached session).
3. **Inline memory injection**: Enable reference-based memory usage — e.g., `--context ` flag or reusable conversation thread ID.

**Benefits**

- **Dramatically reduces token usage** in repetitive workflows
- Enhances usability and developer experience
- Makes Developer Pro subscription more scalable
- Enables more advanced, stateful automation scenarios

**Why this matters**

The current architecture essentially penalizes power users who attempt to use the CLI seriously. Token-based limits are understandable, but **statelessness forces inefficiency**. Adding some form of memory — even limited — would resolve most of these frustrations and bring Amazon Q CLI in line with modern AI development tools.

**Closing Thoughts**

Amazon Q has huge potential. Adding memory would make it a true assistant, not just a command-line responder. I’d be happy to test or provide feedback if this feature is prioritized.

Thank you for considering!