The problem with a single AI personality
For a long time I used LacPointer with the default persona for everything. Morning standup notes, debugging sessions, half-baked shower thoughts I wanted to capture. It worked fine. But "fine" is the wrong bar when you're using something a hundred times a day.
The default persona is warm and helpful. That's exactly what you don't want when you're deep in a gnarly bug and need someone to just tell you what's wrong with your code — no softening, no "great question!", just the answer. And it's also not what you want at 11pm when you're spiraling about a product decision and need something closer to a thinking partner than a coding assistant.
That's what LacPointer Personas actually solve. Not gimmick stuff — a real way to swap the entire brain behind the bar depending on what you're doing.
What a Persona actually is
Each Persona has five parts:
- Name — up to 40 characters, shows up in the bar header when active
- Avatar — optional image, small but it helps you glance and know who you're talking to
- Voice — one of six: alloy, verse, ash, sage, coral, or cedar. This bakes into every voice session automatically
- System prompt — replaces the default LacPointer prompt entirely. This is the most important part
- Greeting — shown when you activate the persona in voice mode with
Cmd+Shift+V
There's also a tools_allowed field. Leave it empty and the persona can use everything — Notion, Slack, GitHub, all your installed Skills. Or lock it down to a specific subset if you want a persona that only does one thing.
The five built-in ones
LacPointer ships five personas out of the box, all read-only but cloneable so you can tweak them:
- Default — the standard friendly assistant. Good for general use.
- Coding — senior engineer mode. Direct, skips the hand-holding, gives you the code first and the explanation only if you ask.
- Therapist — non-judgmental, reflective, slow and warm. Doesn't try to fix things immediately, just helps you think.
- Stoic — Marcus Aurelius / Epictetus energy. When I'm overthinking a decision, this one genuinely reframes it fast.
- Hype — high-energy motivational coach. I use this exactly once a week, usually Monday morning. It's unhinged but effective.
The Coding persona alone changed how I use the bar during work hours. No pleasantries, no filler — you paste a function, it tells you what's broken. If you've been using Default for everything, switch to Coding for a day and see how different it feels.
Building your own
This is where it gets useful. The built-ins cover broad categories, but your workflow is specific. Here's a persona I built that I use constantly:
I called it Spec Writer. The system prompt tells it to respond in structured PRD format — problem statement, proposed solution, open questions, out of scope. Voice set to sage (calm, measured). No greeting, because I usually activate it in text mode. tools_allowed set to just Notion so it can push the spec directly to a page when I'm done thinking out loud.
The workflow: I hit Option+Space, switch to Spec Writer, and just talk through an idea. It asks clarifying questions in PRD framing, which forces me to think clearly about what I'm actually proposing. When I say "push this to Notion," it formats and sends the whole thing. No copy-pasting.
Another one I have: PR Reviewer. System prompt written like a senior engineer doing a code review — focus on edge cases, security, naming, test coverage. No encouragement, just notes. Paste a diff, get back a numbered list of issues. I use this before asking a real human to review anything significant.
Voice mode and personas
This is the part that doesn't get enough attention. When you activate voice mode with Cmd+Shift+V, the active persona's voice and system prompt bake into the entire session. That means your Stoic persona genuinely sounds different from your Hype persona — different voice, different cadence in the responses.
The greeting field is specifically for this. When I switch to my Spec Writer persona and hit Cmd+Shift+V, the greeting tells me it's ready and what it's expecting. Small thing, but it creates a mental context switch that makes the session more focused.
If you use LacPointer for voice control — navigating your Mac, opening apps, clicking things — the active persona doesn't get in the way of that. Voice commands are handled at a lower layer. The persona shapes the conversational AI layer on top.
A few tips from actual daily use
Clone before you edit. The built-in personas are locked, but you can clone any of them and modify the clone. Starting from Coding and adjusting the system prompt is faster than writing one from scratch.
Keep system prompts short. I've tried elaborate multi-paragraph prompts and they don't work better than a tight four-sentence version. The model reads it on every message, so dense and specific beats long and vague every time.
Use the tools_allowed restriction intentionally. If you have a persona for journaling or thinking out loud, you probably don't want it randomly hitting your GitHub or Slack integrations. Lock it down and the persona stays in its lane.
Switch personas by tapping the persona name in the bar header. One tap, instant switch, no settings menu required. The bar itself changes to reflect who's active — the name updates, the avatar changes if you set one. You can tell at a glance which brain you're talking to.
The practical starting point
If you haven't touched Personas yet, start here: open LacPointer with Option+Space, tap the persona indicator in the header, and switch to Coding. Paste a function you've been looking at for too long. See what comes back. That one switch will probably convince you to spend twenty minutes building a few of your own.
The whole system is sitting there unused for most people. It's one of the more considered features in the bar — give it an actual try before writing it off as cosmetic.