Much like my friend Andre's stance on sovereign creators needing to control the digital dirt they build on, I find that I prefer to be free from the whims of any given company -- especially when it comes to the generative AI side of things.
Why I Rejected Platform "Memories"
Way back when, ChatGPT came out with some interesting features around the platform building "memories" about you and your preferences. Sounded neat, but I immediately confirmed that was disabled in my account. I can immediately point to at least three reasons.
1. Control. I have way too many different areas that I might be exploring with an LLM at any given moment and the LLMs already seem to want to overgeneralize to begin with. I know they'd decide some offhanded comment in a random thread now applies to everything I'm doing on their whole platform. I tend to curate my conversations carefully, so I don't want to hand that control over to the LLM.
2. Transparency. I have no idea what is stored and where it is being used. They might have all the right information, but I wouldn't know it. Even if they give me a user interface to look at what they say is stored, doesn't mean that's all of it. If I manually manage those sorts of "memories" and add them to a conversation manually, I know exactly what is going into that thread. (With the asterisk for the system prompts that the providers control out of sight.)
3. Portability. Anything that ChatGPT might store as a memory for me would vary from what Claude might store or Gemini. If I move to another model for whatever reason, any of those things I have done to tailor my environment for a given provider stays locked in that provider. And having seen how quickly the winds change with big companies, let alone the massive shifts we are seeing in generative AI specifically, I want to be able to easily "take my data with me" as I test different providers and models.
My Manual Context Library
Even before the whole "memory" thing was available on any of the platforms, I was maintaining my own equivalent. If there was something I added to a chat more than a time or two, I started adding it to a text document so I could just grab it and add it to a relevant conversation. I had one about myself, one about my coding preferences, one about Tiny Digital Worlds and Productive Tension and all the other models I lean into in my life. If it was something I came back to, I added it to my little library of snippets.
While I was managing it by hand, I actually think that was a benefit, not a downside. I had to take a moment and consider what else the current chat might benefit from knowing about me and my preferences. I saw what I added and had the chance to edit it.
These days, I have an agent that runs on my computer with those same sorts of files available to it as context. The agent can make tool calls to include files it thinks may be relevant or I can explicitly include a file or two. I am much less likely to actually check the individual file contents and tailor it to the task at hand -- and I'm only just realizing that as a cost of the convenience of letting the agent grab the files directly.
The LLM Wiki Approach
A lot of folks have recently joined this idea of having a library of plain text or Markdown files available to the agentic systems. Obsidian is an option for managing the library with a more human-readable interface. (I just use the file browser and a text editor, but I also have them organized in ways that make sense to me, not just an LLM.) If you let the agent organize it for you, the idea of an "LLM wiki" parallels the Agent Skills standard for progressive disclosure.
All of these avoid the overhead of a database, but give up the option for "semantic search" -- searching for similar meaning, not just matching words. If you aren't doing this for a large corporate collection of content, you're probably fine with the plain text route.
What Would Lock Me In?
I've been asked what it would take to "lock me in" to a given provider. I honestly can't think of anything in particular. There are features I'll test on a given provider and I also know that another provider will have something similar soon enough or an open source software project will implement it so it works across providers.
I saw Claude CoWork controlling your computer and OpenClaw accessing all your digital accounts and said "no thank you!" I can see the convenience factor, which is why I did eventually start testing Hermes Agent, but I'm not willing to give over the keys to the castle in an experimental tool or vibe coded platform with known security issues. I did hear how someone was using one of these systems in an experiment when their personal assistant moved away, so I see how it can be useful. But being who I am, I'm way too tempted to just create my own or extend an open source system. (Which is no longer limited to just geeky folks like myself -- if you have the idea and the tokens to spend, an awful lot of personal utility is right there. Relying on vibe coded tools for your business is a whole 'nother beast and a topic for another day.)
Spreading My Bets
Heck, I'm not even committed to one provider's models. I am always on the lookout for new providers that take on a different angle. Cerebras for ridiculously fast responses. Blackbox for double credits -- API and cloud usage each credited what you're paying each month. NeuralWatt for energy-based pricing, which can be ridiculously cheap compared to token-based billing. Along with the direct providers and the aggregators like OpenRouter and OpenCode Zen/Go and KiloCode. I would rather throw my $20 per month at various providers than one of the 800-pound gorillas and see if someone has that angle that might convince me to switch over.
These are all little ways I keep my independence from any one platform or company being bought by someone I'd rather not do business with, or being used for mass surveillance, or just changing their offerings enough that the value isn't there anymore. (Those aren't arbitrary examples. Each one has happened here in the first four months of 2026.)
I find the flexibility and freedom to move around worth the minor inconveniences. You have to make your own decisions on that front. I'd just prefer you do it with your eyes wide open and your needs and preferences clearly in mind.