Meet Personas: Custom AI Characters That Remember Who They Are

After the birth story, on-device AI, and the Connect guide, it is time for the fun stuff. Personas are one of the features people ask for most in Support and Feature Requests, and honestly, I get it. Typing "you are a helpful assistant who speaks like a grouchy pirate" at the start of every chat gets old by Tuesday.

Personas fix that. They are saved characters you build once and reuse whenever you want that personality back.

What is a Persona, in plain English?

A Persona is a profile for your AI: a name, an avatar, a system prompt (who they are and how they talk), and optional extras like a preferred model, voice, memories, even a ComfyUI workflow if you are into image generation.

Think of it less like "settings" and more like casting a character in a show. Same LM Mini app. Same local/private stack. Different role every time you tap a different Persona.

Generic chat is fine for quick questions. Personas are for when you want consistency. The tone sticks. The rules stick. You are not re-explaining yourself every session.

Why I built them

Early LM Mini was one chat window and a pile of knobs. Power users loved it. Everyone else asked: Can I save this personality? Can I have a "work mode" and a "creative mode"? Can my AI remember my project without mixing it into everything else?

So Personas grew into profile pages, not just a text field hidden in advanced settings. You can see who you are talking to. Swap them from a new chat. Attach the right voice. Keep memories scoped to the ones that should know your stuff.

That community feedback loop again. Same story as voice chat and Kokoro. Someone asks, I ship, we iterate.

What you can attach to a Persona

I will not drown you in every toggle (the app docs exist for that), but the highlights:

• Name and avatar. So it feels like someone, not a blank text box.
• System prompt. The soul of the Persona: tone, expertise, boundaries, format ("always reply in bullet points," "never use emojis," whatever you need).
• Model choice. Your tutor might run a small fast on-device model. Your code reviewer might use the big one on your desktop GPU.
• Voice (Kokoro). Different Personas can have different voices and speeds. Useful for voice chat when you want the "character" to sound the part.
• Memories. LM Mini's memory system can remember facts across chats. You can share specific memories with a Persona, or keep a Persona private so its chats do not feed the global memory pool. Handy when work stuff and personal stuff should not mingle.
• ComfyUI workflow (if you use image gen). Link a saved workflow to a Persona so your "art director" character and your "quick sketch" character behave differently.

There is also an AI Persona Generator if you want a starting point instead of staring at an empty prompt box. Edit whatever it suggests until it sounds like yours.

Real ways people use Personas

You do not need to be a novelist or a dungeon master. Some examples I see in the wild:

• Writing partner who knows your blog voice and pushes back on fluffy drafts.
• Study tutor who explains things simply and quizzes you at the end.
• Code reviewer who is picky about readability and security, not just "looks fine."
• Brainstorm buddy for product ideas, zero corporate buzzword tolerance.
• Character for stories or roleplay because yes, people do that, and they want it local and private.

Same phone. Same LM Mini. Tap a different face, get a different room.

How to create your first Persona (rough guide)

1. Open LM Mini → Personas (or Persona library / profile section, depending on your app version).
2. Tap Create or use the Persona Generator if you want a head start.
3. Give it a name and pick an avatar so you recognize it later.
4. Write the system prompt: who they are, how they speak, what they should never do.
5. Optional: pick a model, voice, memory sharing rules, image workflow.
6. Save. Start a new chat with that Persona selected.

Send one message. If the reply sounds wrong, tweak the prompt. Personas are cheap to experiment with. You are not reinstalling anything.

Personas vs Group Chat (quick tease)

Personas are one character per chat (usually one model wearing one mask). Group Chat is the writers' room: multiple Personas or models in the same thread, talking to each other or to you. I will cover that properly in a later post. For now: build two or three Personas you like. Group Chat becomes much more fun once you have a cast.

Privacy note

Personas do not change the privacy story. If you are on on-device or local LM Studio / Ollama, the conversation still runs on your hardware. Memories are stored locally unless you use optional Pro cloud backup for chats. Persona-level memory toggles just give you finer control over what gets remembered where.

The bottom line

LM Mini was never meant to be one generic assistant bolted onto a model picker. Personas are how you make local AI feel personal without giving up private.

Build one this week. A calm editor, a chaotic idea generator, a patient teacher. Whatever matches how you actually use the app.

Next post in the series: Group Chat, where multiple Personas share one conversation. Until then, go make someone interesting.