The same AI, different modes
There's a version of your AI assistant that gives you short, direct answers with zero fluff. There's another version that asks clarifying questions, slows down, and reflects things back to you. A third one explains concepts like you're a junior dev who needs the full picture.
These aren't different tools. They're all LacPointer, just with different Personas active.
I used to deal with this by writing a long system prompt, pasting it at the start of a new chat, and hoping the model held the tone for more than three messages. It never really did. Personas solve this properly — the system prompt bakes in at the session level, not at the message level.
What a Persona actually is
A Persona in LacPointer is a named AI personality with five configurable pieces:
- Name — up to 40 characters, shows up in the bar UI
- Avatar — optional image, makes it easier to know at a glance which mode you're in
- Voice — one of six: alloy, verse, ash, sage, coral, or cedar. This voice carries into every voice session with that persona active
- System prompt — replaces the default LacPointer prompt entirely. This is the most important field.
- Greeting — the message shown when you activate the persona in voice mode. Small detail, but it matters for getting your head in the right space.
You can also set tools_allowed to restrict which skills and integrations the persona can reach. Leave it empty and it has access to everything. Narrow it down and you can build a focused assistant that only touches specific APIs.
The five that ship by default
LacPointer comes with five built-in personas. You can't edit them directly, but you can clone any of them as a starting point for your own.
- Default — the general assistant. Friendly, wide scope, no particular slant.
- Coding — senior engineer mode. Direct, no hand-holding, assumes you already know the basics. This is the one I have open for most of my day.
- Therapist — non-judgmental, reflective, warm. Useful when you're thinking out loud and want something that listens more than it answers.
- Stoic — calm, philosophical, leans on Marcus Aurelius and Epictetus. I pull this one up on rough days more than I'd like to admit.
- Hype — high energy motivational coach. It's exactly what it sounds like. I use it before giving talks.
All five are read-only, which makes sense. You don't want to accidentally break your baseline.
Building your own
This is where it gets useful. The built-ins are convenient, but my most-used persona is one I built: a technical writer mode I call Precise.
The system prompt is something like: "You are a technical documentation editor. Every response should be concise, accurate, and free of filler. Prefer short paragraphs. When asked to explain something, explain it once, clearly. Do not offer alternatives unless explicitly asked."
I gave it the ash voice because it's calm and measured. I set its greeting to something minimal. And crucially, I left tools_allowed empty — Precise gets everything: Notion, GitHub, Calendar, whatever skills I have installed. When I'm drafting docs, I might need to pull a GitHub issue or push something to Notion, so I want full access.
Creating a persona takes maybe two minutes. You fill in the fields, hit save, and it's immediately available from the bar. One tap to switch.
Voice sessions inherit the persona
This is the part that surprised me when I first tested it. When you press Cmd+Shift+V to start a voice session, the active persona's voice and system prompt carry straight into it. You don't have to do anything extra.
So if I switch to my Coding persona and then hit Cmd+Shift+V to ask something verbally about a function I'm looking at, the response I get back is in the Coding persona's voice — literally, the audio voice — and it answers in that terse, senior-engineer style. It's coherent in a way that copy-pasting a system prompt into a chat never really was.
The Therapist persona uses the coral voice by default. It's noticeably softer than ash or cedar. That kind of alignment between personality and audio voice is a small thing, but it makes the interaction feel intentional rather than like you bolted AI onto a voice API.
Restricting tools on purpose
I have one persona called Focus. It has a system prompt that's essentially: "You are a distraction-free writing assistant. Help the user write and edit text. Do not discuss anything unrelated to the document at hand."
For this one, I actually use tools_allowed to lock it down. No GitHub, no Crypto Tracker, no Hacker News skill. Just the bare assistant. When Focus is active, LacPointer can't go fetch a trending repo or check my calendar even if I ask. That constraint is the point.
This is what makes Personas more than a cosmetic feature. You're not just changing the tone — you're changing what the assistant is allowed to do.
Cloning an existing persona
If you want to start from one of the built-ins, clone it first. The clone becomes a fully editable copy. I cloned Coding and tweaked the system prompt to add context about our specific stack — TypeScript, Supabase, Fastify. Now when I ask it about auth middleware it doesn't suggest Express patterns by default.
Stack-specific context in a persona is something I wish I'd set up earlier. It cuts down on the back and forth where you're correcting the model's assumptions every other message.
Switching between them
Hit Option+Space to open the bar, then tap the persona name or avatar. One tap, instant switch. The active persona is always visible so you know what mode you're in.
I switch probably five or six times a day. Morning is Coding. Mid-afternoon when I'm writing is Precise. If I'm stuck on something and want to think through it differently, I'll flip to Stoic for a few minutes. It sounds more complicated than it is — once it's part of the muscle memory it's no different from switching a browser tab.
One practical tip
Write your system prompts in the second person, not the third. "You are direct and concise" works better than "This assistant is direct and concise." Models respond to second-person framing more reliably, especially when the system prompt is long. Keep it under 300 words if you can — tight prompts tend to hold their character across a long session better than sprawling ones.
If you haven't set up a custom persona yet, clone Coding, add a line or two about your actual stack and working style, and use that for a week. It's one of those features that's hard to justify on paper and then impossible to go without once you're used to it.
Download LacPointer at lacai.io/download and find the Personas section in settings.