Multi-Agent Orchestration Stack

AI-Guided Setup Prompt — Multi-Agent Orchestration Stack

Paste this entire prompt into any AI tool (Claude, ChatGPT, Cursor, Gemini) to get step-by-step guidance building a local multi-agent system with zero API cost.

---

REFERENCE IMPLEMENTATION:

Verified stack components:

• crewAI: repoverifier.dev/reviews/crewaiinc-crewai
• ollama: repoverifier.dev/reviews/ollama-ollama

---

You are a senior developer helping me build a local multi-agent orchestration system using the Multi-Agent Orchestration Stack verified at repoverifier.dev/solutions/multi-agent-orchestration-stack.

I want to build a crew of AI agents that collaborate on tasks — each agent has a role, gets assigned tasks, and hands off results to the next agent — fully local, zero API cost, no cloud services, no API keys required.

Guide me one step at a time. Wait for my confirmation before moving to the next step. If something fails, help me debug it before continuing.

---

STACK:

• ollama — runs the LLM locally (llama3.2 model)
• crewAI — orchestrates multi-agent crews with role assignment, task delegation, and context passing

Both components have been independently verified as SOLID at repoverifier.dev. Tested end-to-end on macOS Apple Silicon with Python 3.13.

---

KNOWN GOTCHAS — warn me about these before each relevant step:

1. CLI not in PATH: after pip install crewai, the crewai command

is not found by default on macOS. Fix with:

export PATH="/Users/$(whoami)/Library/Python/3.13/bin:$PATH"

Add this to your ~/.zshrc to make it permanent.

2. Default model mismatch: crewai create crew scaffolds with

ollama/llama3.1 by default. If you don't have llama3.1 pulled,

the first run fails. Check ollama list and update .env to match.

3. Dependency conflicts: crewAI overwrites openai and rich versions.

Always use an isolated venv — never install into system Python.

4. Python version: crewAI requires Python >=3.10 <3.14.

Python 3.13 works correctly.

5. uv required: crewai run uses uv internally to manage the project

venv. It will be installed automatically if missing.

---

STEP 1 — Verify prerequisites

Ask me to run these commands and share the output:

python3 --version

ollama --version

ollama list

Required:

• Python 3.10–3.13
• ollama installed and running
• At least one model in ollama list (llama3.2 recommended)

If ollama is missing: go to https://ollama.com and install it.

If llama3.2 is missing:

ollama pull llama3.2

Do not move to Step 2 until all prerequisites pass.

---

STEP 2 — Install crewAI

Tell me to run:

pip install crewai

Then fix the PATH issue:

export PATH="/Users/$(whoami)/Library/Python/3.13/bin:$PATH"

crewai --version

Expected: crewai, version 1.14.x

Warn me: add the export line to ~/.zshrc to avoid repeating this every session.

---

STEP 3 — Scaffold a crew

Tell me to run:

mkdir my-crew && cd my-crew

crewai create crew my_crew

When prompted:

• Select provider: ollama (option 7)
• Select model: ollama/llama3.1 (option 1)

Then immediately fix the model to match what's installed:

cd my_crew

sed -i '' 's/ollama\/llama3.1/ollama\/llama3.2/g' .env

cat .env

Expected .env:

MODEL=ollama/llama3.2

API_BASE=http://localhost:11434

---

STEP 4 — Run the crew

Tell me to run:

crewai run

Expected: two agents run sequentially — Researcher then Reporting Analyst. report.md is created in the project root.

Tell me to verify:

ls -la

cat report.md

If model not found error: check ollama list and update MODEL= in .env to match exactly.

---

STEP 5 — Verify the handoff

Tell me to open report.md and confirm:

• The file exists and has content
• The content reflects what Agent 1 produced (even if Agent 1 refused or gave a limited answer)
• Agent 2 expanded on Agent 1's output

If both agents ran and report.md has content, the stack is working correctly.