Build your first
AI agent.

Go to ollama.com/download ↗, choose your operating system, install the app, and open it. On a Mac, move Ollama to Applications when prompted. Keep the app running.

Use macOS Sonoma 14 or newer, or Windows 10 22H2 or newer. Apple silicon Macs support GPU acceleration; Windows performance depends on your hardware.

ollama --version

My video uses a MacBook Pro M3 and Qwen 3.5 with 9 billion parameters. Before starting it, set the context length below.

Set the context length: in Ollama’s app settings, increase the context window. Ollama recommends 64K for agent tasks, if your memory allows it. OpenClaw’s automatic discovery needs at least 16K. If memory is tight, try qwen3.5:4b and smaller tasks; expect lower reliability.

Now run this command, wait for the download, and ask the model a simple question:

ollama run qwen3.5:9b

The download is roughly 7 GB. Running it needs additional memory, especially with long conversations. Your available RAM matters more than the laptop’s name.

Model details ↗ Context & memory ↗

Open the official installation guide ↗ and review the installer. Then run the command for your operating system in the second terminal window. The installer handles runtime setup and starts onboarding.

curl -fsSL https://openclaw.ai/install.sh | bash

iwr -useb https://openclaw.ai/install.ps1 | iex

If installation finishes without opening onboarding, run:

openclaw onboard

Choose One agent if asked, and skip optional cloud memory embeddings. Review the permissions during setup. For this first agent, keep the Gateway on its default local interface and skip optional messaging integrations.

In onboarding, choose the custom/manual connection path if needed, then use these settings. Labels may vary slightly with your installed version.

Provider

Ollama

Mode

Local only

Ollama host

http://127.0.0.1:11434

Default model

qwen3.5:9b

Choose the model without a :cloud suffix. Finish onboarding, then launch the browser dashboard:

openclaw dashboard

Keep the Gateway terminal open if setup launched it in the foreground. If your model is missing, check that Ollama and the model are running, then consult the Ollama connection guide ↗.

Paste this into the dashboard chat. Replace the brackets with your details first, then check that Optimus saves the preferences to its workspace files.

Your name is Optimus. You help me with research, lead generation, and sales and marketing drafts.

My business: [what you sell and who you help]
My ideal client: [industry, location, size, and problem]
My preferred writing style: [tone and language]

Cite your sources. Separate facts from assumptions. Mark missing information as unknown. Never invent contact details.

Ask before sending messages, spending money, or changing files outside the task. Save these approved preferences to your workspace identity and user files so you can use them in future sessions.

You can revise these preferences later. Specific context—your offer, customer, and definition of a useful result—makes the agent more useful than a name alone.

For a free option without an API key, install OpenClaw’s official Parallel plugin, then open web configuration. Use a new terminal window if the Gateway is occupying your current one, and run these commands one at a time:

openclaw plugins install @openclaw/parallel-plugin

openclaw configure --section web

Select Parallel Search (Free)—provider ID parallel-free. In the Control UI, open Settings → Search and test it. Start a new chat after configuring the tool.

Official free-search setup ↗

Download the lead-criteria.md template ↓, fill in the brackets in a text editor, and put a copy in OpenClaw’s workspace. The default location is ~/.openclaw/workspace; use your configured workspace if you changed it.

On Mac, use Finder → Go → Go to Folder and enter that path. On native Windows, use %USERPROFILE%\.openclaw\workspace in File Explorer. If you installed through WSL, use the workspace inside WSL.

# Lead criteria

## My business
I help [type of customer] achieve [outcome] through [service or product].

## Find businesses that match
- Industry: [your target industry]
- Location: [country or city]
- Company size: [size range, if relevant]
- Qualifying signals: [specific, publicly verifiable signs of need]
- Exclude: [businesses or categories to avoid]

## Research rules
Use public business information only. Cite the source for every claimed qualification. Mark anything you cannot verify as unknown. Do not guess email addresses or contact anyone.

This gives Optimus the relevant document from your Documents folder without asking it to inspect everything. File access depends on its tool permissions and sandbox configuration; the workspace path alone does not enforce isolation.

Paste this prompt into a new chat. If Optimus cannot read the brief, confirm its actual workspace and file-tool permissions, or paste the brief into the chat.

Read lead-criteria.md in your workspace. Find five businesses on the public internet that match the criteria.

For each business, include:
1. Company name and website
2. Location
3. Evidence that it fits my criteria
4. A publicly listed business contact or contact-page link
5. Source URLs
6. One personalized outreach angle

Do not invent email addresses. Mark unverified details as unknown. Show me the table first. After I approve it, save the result as prospect-research.csv in your workspace. Do not contact anyone.

A small local model can miss details or struggle with tools. If it stalls, ask for one company at a time and keep the output simple.