You've built your first Superagent. Congrats. Now the real work starts: making it do exactly what you want.
Most people fill in the name, hit save, and wonder why their Superagent sounds robotic or misses instructions. The secret is the Soul section of the Brain. That's where your Superagent learns how to think, what to care about, and what lines never to cross.
This guide teaches you the PACE Framework. It's a simple way to structure your Soul instructions so your Superagent actually works the way you need it to.
The Three Parts of the Brain
Before we talk about PACE, let's map out where everything lives in Base44.
Open your Superagent and click the Brain tab. You'll see three sections:
Identity: Name, personality vibe, avatar. This is your Superagent's face.
Soul: The behavioral rules, constraints, and tone. This is the magic section most people skip. It's where you tell your Superagent how to act, what to care about, and what it absolutely cannot do.
Knowledge: Files, FAQs, documents, context. Upload your pricing doc, FAQ, product specs, or customer handbook here. Your Superagent reads it instantly. No training needed.
The Identity gets seen. The Soul makes it work. The Knowledge makes it powerful.
The PACE Framework: Build Your Soul Section
PACE is four simple parts. Write them in order. Your Superagent becomes more aligned with every sentence.
P: Purpose
One sentence. What is this Superagent's core job?
Examples:
- "You are a customer support agent that handles product questions via email and WhatsApp."
- "You are a lead qualification agent that screens inbound sales inquiries and schedules calls for qualified leads."
- "You are a social media manager that responds to comments and DMs on Instagram within 2 hours."
Write it as a direct statement. Your Superagent will live this sentence.
A: Audience
Who will interact with this agent? What do they want? What's their mood?
Examples:
- "You talk to customers who are frustrated or confused. They're not reading 500-word answers. Be concise, empathetic, and solve their problem in 2-3 sentences."
- "You interact with potential clients who are evaluating whether to hire us. They want proof we're competent and that we understand their specific situation. Be warm but professional."
- "You engage with social media followers during working hours. They're scrolling quickly. Your tone should be casual, friendly, and witty without trying too hard."
This section keeps your Superagent from sounding like a bot to the wrong audience.
C: Constraints
What must your Superagent absolutely never do? This is your safety net.
Write 3-4 hard rules. Be specific.
Examples:
- "Never discuss pricing without my explicit approval first. If someone asks for a quote, say you need to check with the team and you'll get back to them."
- "Keep every response under 100 words. Err on the side of short."
- "If you're unsure about something, never guess. Say 'I'll check with the team and get back to you.' Do not make up information."
- "Always use the person's first name when you have it. It makes interactions feel personal, not automated."
- "Never promise a deadline I haven't confirmed. If you don't know when something ships, say 'I'll check our roadmap.'"
These constraints keep your Superagent from breaking things while you sleep.
E: Examples
Show, don't tell. Give your Superagent 2-3 examples of exactly how you want it to respond.
Format:
User says: [realistic input]
You respond: [the response you want — exact tone, length, content]
Example:
User says: "I've been waiting 2 weeks for my order and it's not here."
You respond: "I'm really sorry — that's frustrating. Let me look into this right now. Can you give me your order number? I'll find exactly where it is and let you know what's next."
User says: "Do you have a free trial?"
You respond: "We do — 5 messages per day to test it out. That's usually enough to see if it works for you. Want me to set one up?"
User says: "What's your secret sauce?"
You respond: "We connect to your real tools (Gmail, Slack, Calendar, etc.) so your agent isn't just a chatbot — it works like a team member. Most agents sit in a chat box. Ours integrates everywhere you already work."
Examples are powerful because your Superagent learns your voice, your length, your priorities from real interactions, not abstract rules.
