I have been using Gemini CLI (Command Line Interface) for a while now, and while it is an incredible tool for coding, I have pivoted to using it for my personal life too.

My setup:

A deep memory graph using https://github.com/Beledarian/mcp-local-memory and a series of markdown files that track everything. This includes health, finances, travel etc. As a hybrid approach for a more robust history, I also store every conversation verbatim using https://github.com/mempalace/mempalace .

I have a persona-driven system with strict mandates on tone, ethics, and unfiltered communication. They are not a ‘helpful assistant’, but sharp and opinionated while knowing my history.

We’ve developed custom procedures. This includes a UK-specific retirement modeller, a book recommender that leverages my reading history, local image generation…

Because the agent has access to my full media history, recommendations are really great.

The Workflow:

I spend a lot of time walking my dog, and during this I'll often dictate to my Apple Watch if a thought comes into my head I want to discuss. I have a ubiquitous interface (Watch, Phone, E-Reader) via a Telegram bridge. On the Watch, I use a complication to dictate; if I’m wearing AirPods, Siri reads back the response, otherwise they arrive as notifications.

When in my study or at work, I use the CLI directly in a split screen. I’ll paste context straight into the terminal or tell it to look at what I'm looking at. The system updates the local memory graph with key details so I do not have to repeat myself in future sessions. In-between, I use the Telegram bridge on my phone or Android e-reader. If I finish a book, am thinking about a purchase, or have a travel idea, I send a quick chat message and the memory manager skill updates my tracking files in the background.

It goes beyond media. I recently used the agent to help with a very complex business negotiation that it developed a detailed memory for. It also handles my professional work - creating and supporting design patterns for a large organisation. I take the lead, but the AI reviews my work and suggests improvements or rewrites.

Why the CLI over the Consumer App?

Consumer AI apps have ‘Memory’, so why go to the trouble of building a CLI-first architecture?

Reasoning & Cost: Using the CLI means I am interacting with the raw models via API. I find the reasoning capabilities and adherence to complex instructions (like my retirement modelling) are significantly higher vs the consumer apps. I defaulted to Gemini because it performs well for my needs and is cost-effective - I already had a GeminiPro account and it saves me needing a separate Nest subscription. While I’ve used Claude and ChatGPT, I find OpenAI less appealing as a company and Anthropic's rate limiting can be an issue unless you're spending much more.

Data Ownership: My data is stored in portable, local Markdowns and a standard database. If I decide to switch to a different LLM provider next week, I can. I am…

为什么值得关注

能改变理解方式，而不只是重复常识；符合当前抓取需求；它提供了新的理解或解释，而不只是表面观点

来源：reddit，领域：tech，保留分：0.68