How to Configure Ollama and OpenClaw Locally: Browser Control, Telegram, and Best Models for Mac

Running AI agents entirely on your local hardware gives you privacy, zero API costs, and total control over your system. With the integration between Ollama and OpenClaw, getting a highly capable local AI assistant up and running has never been easier.

In this post, we'll walk through how to launch OpenClaw with Ollama, set up browser control, connect it to a Telegram channel, and choose the right local models based on your Mac's RAM.

---

1. Launching OpenClaw with Ollama

Ollama has made it incredibly simple to run local LLMs, and now it officially supports launching OpenClaw directly. You don't need to manually configure API keys or base URLs — Ollama handles the handshake.

The One-Command Setup

If you already have Ollama installed, simply run:

ollama launch openclaw

Here's what happens under the hood:

1. Installation: If you don't have OpenClaw installed yet, Ollama will prompt to install it via npm.
2. Security Notice: On the first launch, Ollama shows a security notice explaining the risks of giving an AI agent tool access. Read it carefully.
3. Model Selection: You'll be prompted to pick a model from a selector — local or cloud.
4. Onboarding: Ollama configures the provider, installs the Gateway daemon, sets your model as the primary, and automatically installs the web search and fetch plugin (giving your local model the ability to search the web).
5. Gateway Startup: The Gateway starts in the background and the OpenClaw TUI opens.

Once started, your local agent is ready to work.

Important: Local models require a context window of at least 64k tokens to work properly with OpenClaw. Ollama handles this for recommended models, but if you're using a custom model, make sure to set the context length appropriately. See Ollama's context length docs for details.

Non-Interactive (Headless) Mode

For scripts, CI/CD, or Docker:

ollama launch openclaw --model qwen2.5-coder:14b --yes

The --yes flag auto-pulls the model, skips all interactive selectors, and requires --model to be specified.

Changing the Model Without Relaunching

To switch models without starting the gateway and TUI:

ollama launch openclaw --config

Or specify a model directly:

ollama launch openclaw --model kimi-k2.5:cloud

If the gateway is already running, it restarts automatically to pick up the new model.

Web Search for Local Models

The web search plugin is installed automatically when you launch via Ollama. If you need to install it separately:

openclaw plugins install @ollama/openclaw-web-search

Note: Web search for local models requires signing in to Ollama first: ollama signin. Cloud models handle this automatically.

---

2. Configuring Browser Control

For most macOS users, the browser setup you want is the built-in user profile. It lets OpenClaw attach to your normal Google Chrome session through Chrome DevTools MCP, so the agent can use tabs and logins you already have open.

For this setup:

• do not set cdpUrl
• do not launch Chrome with --remote-debugging-port
• do not use the extension or relay flow

This path uses Chrome DevTools MCP, not the managed Playwright browser.

macOS Setup

1. Open your normal Google Chrome.
2. Open chrome://inspect/#remote-debugging.
3. Turn on remote debugging.
4. Keep Chrome open.
5. If Chrome asks for permission, approve the attach prompt.

Then verify that OpenClaw can see your live browser session:

openclaw browser --browser-profile user start
openclaw browser --browser-profile user status
openclaw browser --browser-profile user tabs

What success looks like:

• status shows driver: existing-session
• status shows transport: chrome-mcp
• tabs shows your current Chrome tabs

If attach does not work:

• make sure remote debugging is still enabled at chrome://inspect/#remote-debugging
• make sure Google Chrome is still open
• restart the OpenClaw Gateway and try again

When you want the agent to use your signed-in Chrome session, use profile="user".

Warning: The user profile gives the agent access to your signed-in browser session. Only use it when you're at the computer and can monitor what it is doing.

---

3. Connecting OpenClaw to Telegram

If you want to chat with your local OpenClaw setup from your phone, connect it to Telegram. For most people, a simple DM bot is enough.

1. Create a bot with @BotFather and copy the token.

2. Configure Telegram in OpenClaw. The easiest way is:

openclaw configure --section channels

If you prefer editing config directly, the minimum setup is:

{
  channels: {
    telegram: {
      enabled: true,
      botToken: "123456:ABC-DEF1234ghIkl-zyx57W2v1u123ew11",
      dmPolicy: "pairing"
    }
  }
}

3. Start the gateway:

openclaw gateway

4. Message your bot in Telegram. The first time, approve the pairing code:

openclaw pairing list telegram
openclaw pairing approve telegram <CODE>

That is enough for a private Telegram DM setup. Group policies, allowlists, and forum-topic routing are advanced options and do not need to be part of your first local setup.

---

4. Which Model Works Best on Mac?

Choosing the right model is crucial for local performance. If a model is too large, it will swap to disk and run incredibly slowly. If it's too small, the agent won't be smart enough to use tools or browser control effectively.

Mac Hardware Matrix

Here is our recommended matrix for MacBooks and Mac Minis running Apple Silicon (M1/M2/M3/M4). Remember: local models need at least 64k context to work properly with OpenClaw.

Mac RAM	Recommended Model	Approx. Size (4-bit)	Use Case & Performance	Command
8 GB	qwen2.5-coder:7b	~4.7 GB	Basic coding and chat. Browser control might struggle with complex pages. Tight on memory.	ollama pull qwen2.5-coder:7b
16 GB	qwen2.5-coder:14b	~9.0 GB	Sweet Spot. Leaves enough RAM for macOS and browser. Excellent tool use and reasoning.	ollama pull qwen2.5-coder:14b
16 GB	deepseek-r1:14b	~9.0 GB	Great for deep reasoning tasks, slightly slower output than Qwen.	ollama pull deepseek-r1:14b
32 GB	glm-4.7-flash	~25 GB	Ollama's recommended local model. Strong reasoning and code generation.	ollama pull glm-4.7-flash
32 GB+	qwen2.5-coder:32b	~20 GB	Desktop-class power. Competes closely with cloud models for coding and browser tasks.	ollama pull qwen2.5-coder:32b
64 GB+	llama3.3:70b	~43 GB	Ultimate local intelligence. Requires serious hardware but offers world-class performance.	ollama pull llama3.3:70b

If your Mac does not have enough RAM for a good local model, use a cloud model with the same workflow:

ollama signin
ollama launch openclaw --model kimi-k2.5:cloud

Wrapping Up

The shortest path is simple: install Ollama, run ollama launch openclaw, pick a model that matches your Mac, and only add browser or Telegram setup if you actually need those capabilities.

For most people:

• start with ollama launch openclaw
• use qwen2.5-coder:14b on a 16GB Mac or glm-4.7-flash on a 32GB Mac
• use profile="user" only when you need your real signed-in Chrome session
• add Telegram only if you want to chat with OpenClaw from your phone