OpenClaw + Ollama Local Setup Guide (2026)

OpenClaw (formerly Clawdbot / Moltbot) is a powerful local AI agent framework that connects messaging apps like Telegram and WhatsApp, allowing you to invoke local large models directly from your phone to perform tasks (coding, controlling computer, searching files, etc.).

By pairing it with Ollama, you can achieve a completely offline, free, and API-cost-free setup.

Two Recommended Installation Methods

Method 1: The Easiest Way (Official Recommendation)

Best for: Mac, Linux, Windows (WSL)

Install Ollama (if you haven't already)
- Official Site:
- After downloading and installing, run the following in your terminal (using a 128k+ context model is recommended):
bash
```
ollama pull qwen2.5-coder:32b-instruct     # Recommended, strong tool calling# ORollama pull deepseek-r1:32b# ORollama pull llama3.3:70b                   # If you have powerful hardware
```
Launch OpenClaw directly with ollama launch
bash
```
ollama launch openclaw
```
This command will:
- Automatically install the latest OpenClaw
- Start the onboarding wizard
- Automatically select Ollama as the backend
- Let you choose the model (the one you just pulled)
- Configure Telegram / WhatsApp channels (can be skipped initially)
Once done, it will automatically run the gateway in the background (default port 18789).
Start Using via Telegram
- If you chose Telegram during onboarding, follow the instructions to create a bot and paste the token. Your bot is now ready!

Method 2: Manual Installation (For Advanced Control)

Install Node.js (v22+ recommended)

One-Click Install OpenClaw

Mac/Linux:

bash

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

Windows (PowerShell):

powershell

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

Run Onboarding

bash

openclaw onboard# Or to install as a system service:openclaw onboard --install-daemon

Configure Ollama Backend

In the wizard, select Quick Start -> Skip Cloud -> Select Ollama.
Or manually edit the config file (~/.openclaw/openclaw.json):

json

{  "models": {    "providers": {      "ollama": {        "baseUrl": "http://127.0.0.1:11434/v1",        "apiKey": "ollama",        "auth": "api-key"      }    },    "default": "qwen2.5-coder:32b-instruct"  }}

Start OpenClaw

bash

openclaw start# Or run in background:openclaw start --daemon

Troubleshooting FAQ

Model lacks tool calling ability?
- Switch to models that support function calling well, such as qwen2.5-coder, qwen3, deepseek-r1, llama3.3.
- We recommend at least 14B+ models; 8B models may hallucinate tool calls or forget context.
Too slow?
- Try smaller models (e.g., qwen2.5-coder:14b or 32b).
- Or rent a GPU pod on RunPod / Vast.ai to run Ollama.
Best model for Chinese?
- Current (Feb 2026) recommendation: qwen3 series > deepseek-r1 > llama3.3-Chinese.
No response on Telegram?
- Check if OpenClaw is running (check terminal logs).
- Check if Ollama is running (ollama ps).
- Re-run ollama launch openclaw to reconfigure.

Basically, ollama launch openclaw is the most hassle-free way right now. Most people can get it up and running in 5-10 minutes.