The problem with one-size-fits-all AI
I use LacPointer for everything — quick Notion tasks, thinking through code problems, grinding through writing, even the occasional "am I overreacting to this Slack thread?" check. The issue is that what I need from the AI changes completely depending on what I'm doing.
When I'm debugging a gnarly async issue at 11pm, I do not want the AI to be warm and encouraging. I want it to be a senior engineer who just tells me what's wrong. When I'm trying to untangle something personal mid-afternoon, I don't want terse bullet points. I want something that actually listens first.
That tension is exactly what Personas solve.
What a Persona actually is
A Persona in LacPointer is a configuration object that fully replaces the default AI behavior. It has five parts:
- Name — up to 40 characters, shows in the bar and the persona switcher
- Avatar — optional image, displayed next to responses
- Voice — one of six options:
alloy,verse,ash,sage,coral, orcedar - System prompt — replaces the default LacPointer prompt entirely. This is where the personality lives.
- Greeting — the line spoken when you activate the persona in voice mode (Cmd+Shift+V)
There's also an optional tools_allowed field. Leave it empty and the persona has access to everything — Notion, Slack, GitHub, Calendar, installed Skills, all of it. Lock it down to a specific list and the persona only works with those integrations. Useful if you want a focused assistant that can't accidentally fire off a Slack message when you're just trying to think out loud.
The key thing: whatever voice and system prompt you set, they bake into every single interaction — chat and voice — automatically. You don't have to paste your prompt every session. You don't have to remember to say "talk like a senior engineer." You switch persona once, and the whole bar becomes that persona.
The five built-in personas
LacPointer ships with five system personas. They're read-only, but you can clone any of them and edit the clone.
Default
The general-purpose one. Friendly, capable, doesn't assume what you're working on. Good baseline when you're doing mixed tasks and don't want any particular slant.
Coding
This is the one I live in most of the day. Senior engineer mode — direct, no hand-holding, no filler. Ask it why your useEffect is firing twice and it'll just tell you, without preamble. It won't celebrate you for writing a for loop. It's the persona I wish I'd had years ago when I was wading through bad Stack Overflow answers.
Therapist
Non-judgmental, reflective, slow. It asks questions before jumping to answers. It doesn't try to fix everything. If you've ever vented to a regular AI assistant and felt like you got a 6-step action plan back when you just needed someone to hear you — this persona is the correction for that.
Stoic
Calm, philosophical. Draws from Marcus Aurelius and Epictetus. Good for when you're catastrophizing something and need a more grounded perspective. I use this one when a PR review stings more than it should.
Hype
High energy, motivational. Not subtle. Sometimes you just need someone to tell you to ship the thing and stop second-guessing it. This persona does that, loudly.
Switching between them
Switching is one tap in the bar. Open LacPointer with Option+Space, hit the persona switcher, pick a new one. If you're in a voice session (Cmd+Shift+V), the new persona's greeting plays automatically so you know the switch landed.
There's no session bleed. The Therapist persona doesn't carry context from your Coding session. Each persona starts clean, which is mostly the right behavior — you don't want your philosophical reflection session contaminated by your earlier argument about tab width.
Building a custom persona
This is where it gets useful for anyone who has specific workflows the five defaults don't cover.
I built one called PR Reviewer. Its system prompt tells it to act like a senior engineer doing a code review — focus on correctness, performance, and edge cases, flag anything unclear, and don't soften feedback. I gave it the ash voice because it sounds appropriately measured. I locked its tools_allowed to GitHub only, so it can pull PR diffs but can't go off and create Notion pages mid-review.
Another one I use is a Writing Editor persona. System prompt: edit for clarity and concision, cut anything that doesn't earn its place, flag passive voice, don't rewrite my voice into something generic. Voice set to sage. No tools allowed — it's purely conversational, I paste text and it responds.
The system prompt field is where you spend your time. A few things I've found that make a big difference:
- Be explicit about what the persona should not do, not just what it should do. "Don't ask clarifying questions before answering" is more useful than "be direct."
- Give it a format preference. "Always lead with the answer, add context after" or "use bullet points only when listing more than three things."
- If you're restricting tools, mention it in the prompt too. "You only have access to GitHub. Don't suggest Notion or Slack actions."
- The greeting matters more than you'd think. In voice mode, hearing the persona announce itself helps your brain actually context-switch. Make it short and specific — something that signals what mode you're in.
The voice piece specifically
Each of the six voices has a distinct feel. Quick breakdown from actually using them:
alloy— neutral, slightly formal, the default if you don't pick anythingverse— warmer, conversational, works well for the Therapist personaash— measured, deliberate, good for analytical personassage— calm and clear, my pick for anything writing-relatedcoral— friendly and a bit lighter in tonecedar— grounded, works well for the Stoic persona
The match between voice and system prompt is subtle but real. If you write a Stoic persona and give it the Hype voice, the disconnect is jarring. It's worth spending five minutes on this pairing — voice is the first thing you hear when you open a session, and it sets the tone before the AI says a single word.
A practical starting point
If you haven't touched Personas yet, start by cloning the Coding persona and modifying the system prompt to match how your team actually works. Add the languages you use, the patterns you care about, anything domain-specific that the generic version doesn't know.
Then build one that's purely for thinking — no tools allowed, no task execution, just a focused conversational space where you can reason through problems without accidentally triggering an integration. I call mine "Rubber Duck" and it's become the most-used persona I have.
Personas are available on both Free and Pro. If you haven't downloaded LacPointer yet, grab it at lacai.io/download — it's a two-minute install and you can get to the persona builder from the bar settings immediately.