The default persona is fine. Custom ones are better.
When I first started using LacPointer, I kept the default persona for weeks. It's a friendly general assistant, it works, it answers questions — no complaints. But I kept noticing something: the tone felt slightly off depending on what I was doing. When I'm deep in a debugging session I don't want warm and friendly. I want terse, direct, and technically precise. When I'm doing a late-night planning session I want something a bit more thoughtful.
The five built-in personas — Default, Coding, Therapist, Stoic, Hype — cover a lot of ground. But they're read-only. You can clone them as a starting point, or you can build your own from scratch. I've done both, and now I have four custom personas I switch between depending on the day.
Here's how the whole thing works and what I've actually built.
What a persona actually controls
Each persona has five moving parts:
- Name — what shows up in the switcher, max 40 characters
- Avatar — optional image, just for visual identification
- Voice — one of six:
alloy,verse,ash,sage,coral, orcedar - System prompt — this fully replaces LacPointer's default prompt. Whatever you write here becomes the AI's identity and instructions
- Greeting — shown when you activate the persona in voice mode via Cmd+Shift+V
- Tools allowed — an optional list. Leave it empty and the persona can use everything. Restrict it and the persona only gets those tools
That last one — tools allowed — is underused and actually pretty powerful. More on that in a bit.
Starting from scratch vs cloning a built-in
If you want to create something genuinely different, start fresh. Hit the Personas tab, click New Persona, and you're looking at a blank form.
If you want to tweak something close to an existing personality — say, the Coding persona but with your specific stack in mind — clone it first. Open the built-in, tap Clone, and you get an editable copy. This is the faster path when the built-in's system prompt is already 80% of what you want.
I cloned the Coding persona once to make a "Reviewer" persona that only does code review, refuses to write new code, and always asks me what I've already tried before giving feedback. Small tweaks to the system prompt, but it fundamentally changed how I use it.
The system prompt is where it actually lives
This is the part people underestimate. The system prompt doesn't just set a tone — it fully replaces LacPointer's default instructions. So if you want the AI to always respond in bullet points, or always ask a clarifying question before answering, or never suggest external tools, that all goes here.
A few things I've found actually work well in system prompts:
- Explicit output format instructions: "Always respond in under 4 sentences unless I ask for more detail." This keeps the floating bar responses tight, which matters because you're reading them in a small UI.
- Context about your stack: "The user works primarily in TypeScript with a Next.js frontend and a Node/Express backend. Default to these when making code suggestions." You don't want to re-specify your stack every session.
- Behavioral rules: "Never start a response with a greeting or affirmation. Get straight to the answer." Stops the AI from saying "Great question!" every time.
- Escalation behavior: "If the user seems stuck, offer one concrete next step — don't just explain the problem back to them."
Keep it specific. Vague instructions like "be helpful" don't change behavior. Specific instructions like "if you mention a terminal command, always wrap it in a code block" actually do.
Voices matter more than you'd expect
I didn't think much about voice selection until I started using voice mode regularly via Cmd+Shift+V. The voice a persona uses bakes into every realtime voice session — it's not just a cosmetic choice.
Some of them are noticeably different in pacing and delivery:
ashandsageare calm and measured — good for anything you're using to think through a problemversehas more energy — better for the Hype-style motivational use casecedaris the one I use for my Reviewer persona — it sounds more deliberate, which fits the critical feedback mode
Try a few before settling. It's one of those things that sounds trivial until you realize you're spending 20 minutes a day talking to this thing.
Locking down tools with tools_allowed
By default a persona can use all your installed skills and integrations — Notion, GitHub, Slack, Calendar, whatever you've set up. Most of the time that's what you want.
But there are cases where restricting tools makes the persona more focused. I have a "Focus" persona I use during deep work blocks. Its system prompt tells it to stay on the current task and not get distracted. And I've set tools_allowed to empty — meaning I've left out my Slack and Calendar integrations so the persona literally cannot pull in notifications or check my schedule even if I ask it to.
It's a small friction that reinforces the intent of the mode. When I switch to Focus persona, I'm not just changing the AI's tone — I'm actively restricting what it can reach.
You can list any combination of your installed skill tools and core integrations. If you're not sure what's available, the persona editor shows you everything that's installed.
The personas I actually use daily
For what it's worth, here's what I've landed on after a few months of tweaking:
- Reviewer — cloned from Coding, system prompt focused on critique and questions. Voice: cedar. No new code generation.
- Focus — minimal system prompt, just task awareness. Tools restricted to exclude Slack and Calendar. Voice: sage.
- Brief — default-style personality but with hard output length limits in the system prompt. For when I just want quick answers and don't want a lecture.
- PM Mode — switched to this during planning and writing sprints. System prompt is product-manager-brained: asks about user impact, prioritization, tradeoffs. Voice: alloy.
The built-in Stoic and Therapist personas I use occasionally too — but I haven't cloned them because they're already well-tuned for what they do.
One practical tip before you start
Write your system prompt in a text editor first, not directly in the persona form. It's easier to revise, and you'll want to revise it — the first version is almost never right. After a few days of actual use you'll find the instructions that do nothing and the ones that matter. Trim the noise, sharpen what works.
Then test it in voice mode at least once before you commit to it. The system prompt plus voice combination creates a different feel than text alone, and sometimes what reads well on screen sounds weird spoken out loud.
If you haven't tried building one yet, open LacPointer, hit Option+Space, go to Personas, and clone the built-in that's closest to what you need. From there it's just a few fields. The whole setup takes under five minutes.