Back to Blog
tutorialgetting-starteddevelopersagents

Building Your First AI Agent Team with A.L.I.C.E.

Rob Sanchez2026-03-236 min read

You've probably used an AI assistant before. You type a question, it answers, you type another question. Each conversation starts fresh. There's no memory of what you worked on last week, no understanding of your codebase, no accumulation of context over time.

A.L.I.C.E. is something different. It's a team of 28 specialized AI agents that work together, remember things, and get better the longer you use them. This guide will get you from zero to your first real interaction in about 10 minutes.

Let's do it.


Prerequisites

You need two things:

  1. OpenClaw installed — OpenClaw is the agent runtime that A.L.I.C.E. runs on. If you don't have it, install it first: openclaw.ai
  2. An API key — A.L.I.C.E. supports Anthropic (Claude), OpenAI (GPT-4o), or a local Ollama instance. Have at least one of these ready.

That's it. No Docker, no infrastructure, no config files to hand-edit.


Step 1: Run the Installer

Open your terminal and run:

npx alice-agents@latest

The installer will ask you a few questions:

Welcome to A.L.I.C.E. — Adaptive Learning Intelligence for Coordination & Execution

? Which AI provider would you like to use?
  ❯ Anthropic (Claude)
    OpenAI (GPT-4o)
    Ollama (local)

? Enter your Anthropic API key: sk-ant-...

? Which tier would you like to start with?
  ❯ Starter (free)
    Pro

? Where should A.L.I.C.E. store your workspace? ~/alice-workspace

Answer the prompts and the installer does the rest. It scaffolds your workspace, initializes all 28 agents with their SOUL files, and sets up the memory system. Takes about 90 seconds.

When it's done, you'll see:

✓ A.L.I.C.E. installed successfully
✓ 28 agents initialized
✓ Memory system ready
✓ OpenClaw integration configured

Your orchestrator is Olivia. Say hello:
  openclaw chat olivia

Step 2: Your First Conversation with Olivia

Olivia is your orchestrator — the agent you talk to directly. She coordinates the other 27 specialists behind the scenes. You don't need to know who does what; just tell Olivia what you need.

openclaw chat olivia
Olivia: Hey! I'm Olivia, your AI orchestrator. I have a team of 27 
specialists ready to help — engineering, research, data, security, 
comms, and more. What are you working on?

You: I'm a solo developer building a SaaS app. I need help across 
the board — code review, writing, research, the works.

Olivia: Perfect. I'll get to know your stack and preferences as we 
work. What are you building?

You: A B2B invoicing tool for freelancers. Node.js backend, 
React frontend, Postgres.

Olivia: Got it. I'm making a note of that for the team. Anytime 
you need something — feature planning, code review, a blog post, 
competitor research — just ask me and I'll route it to the right 
specialist.

Olivia is already learning. That context about your stack just got written to her memory. The next time you ask a technical question, she'll know you're on Node and Postgres without you having to repeat yourself.


Step 3: Assign Your First Real Task

Don't start with a trivial test. Give Olivia something real. The faster you do real work, the faster the agents learn what matters to you.

Try something like this:

You: I need a security review of my authentication flow. We're using 
JWT tokens stored in localStorage, bcrypt for password hashing, 
and we haven't implemented rate limiting yet.

Olivia: On it. I'm looping in Selena (security specialist) 
for a full review. Give me a minute.

[Selena is reviewing...]

Olivia: Selena flagged a few things:

1. **localStorage for JWTs is risky** — XSS attacks can steal 
   tokens. Recommend httpOnly cookies instead.
2. **Rate limiting is a priority** — brute force on your login 
   endpoint is a real threat. Selena recommends express-rate-limit 
   as a quick fix.
3. **bcrypt rounds** — what's your cost factor? Anything below 10 
   should be bumped up.

Want Selena to write up the implementation for the cookie migration, 
or should we start with rate limiting?

That's the loop in action. You talk to Olivia, she routes to the right specialist, you get domain-expert output. You never had to know Selena existed.


Step 4: Understanding Agent Memory

Here's what makes A.L.I.C.E. different from a plain AI assistant: agents remember.

Each agent has three memory layers:

  • SOUL — who the agent is, their personality and values. Set once, doesn't change.
  • AGENTS — operational context: your stack, your preferences, your team structure.
  • PLAYBOOK — accumulated expertise: patterns that worked, things to avoid, domain knowledge specific to you.

After a few real work sessions, Olivia's PLAYBOOK starts filling up. She learns that you prefer direct answers over long explanations. She knows your deploy pipeline. She remembers that you tried a particular approach last month and it didn't work.

You can inspect what an agent has learned:

cat ~/alice-workspace/workspace-olivia/PLAYBOOK.md

You can also edit these files directly. If an agent learned something wrong, fix it. The memory system is just Markdown — no black box, no magic you can't inspect.


Step 5: When to Upgrade to Pro

The Starter tier gives you the full 28-agent framework, local memory, and single-user access. It's genuinely useful and free.

Upgrade to Pro when:

  • You want cloud memory sync — so your context follows you across machines and doesn't live only on your laptop
  • You're adding teammates — Pro unlocks shared agent access and collaboration features
  • You need the Mission Control dashboard — a web UI for tracking agent activity, task history, and what your team is learning
  • You want usage analytics — visibility into what models you're using and what things cost
openclaw upgrade

That opens the browser-based upgrade flow. Takes about 2 minutes.


3 Things to Try Today

Don't overthink it. Here are three concrete things you can do right now to get real value out of A.L.I.C.E.:

  1. Paste in a piece of code and ask for a review. Something you actually wrote, not a toy example. See how the engineering agents handle it.

  2. Ask Olivia to research a competitor. Give her a specific company and ask for a feature comparison or pricing breakdown. Watch her route it to the research specialist.

  3. Describe a decision you're wrestling with. Architecture choice, pricing model, hiring vs. contracting — give Olivia the context and ask what she thinks. You'll get a structured breakdown from the relevant specialists, not just a generic AI answer.

The agents get better the more you use them. Start with something real today and you'll feel the difference by the end of the week.


Questions? Issues? The fastest path to help is our Discord or opening an issue on GitHub. We actually read them.