OpenClaw (MoltBot) Setup Guide: Ubuntu + NVIDIA + Ollama + GLM-4.7 + Telegram

Running a personal AI assistant that never sends your data to a third-party cloud, costs nothing per token, and responds directly in your Telegram chat is no longer a weekend research project — it is a single afternoon of setup. This guide walks you through the complete OpenClaw setup on bare-metal Ubuntu, from NVIDIA driver installation to your first Telegram conversation powered by GLM-4.7 running locally through Ollama.

OpenClaw was renamed from MoltBot in January 2026. The CLI command is now openclaw, and the legacy moltbot alias still works for existing scripts. Throughout this guide we use “OpenClaw” and the openclaw command consistently. If you are arriving from an older tutorial that references MoltBot, the concepts are identical — only the name changed. For a deeper look at what OpenClaw is and why the project matters, see our companion piece on what OpenClaw is and how it differs from standard chatbots.

What Is OpenClaw and Why Run It Locally?

OpenClaw is a personal AI gateway that bridges large language models to the messaging channels you already use — Telegram, WhatsApp, Slack, Discord, and others. Unlike deploying a raw chatbot, OpenClaw manages the gateway process, routes messages through a configurable agent layer, maintains session context, and handles channel authentication. Think of it as the middleware that sits between your LLM backend and your phone.

Running it locally with Ollama and GLM-4.7 has three concrete advantages over cloud-API setups:

This approach makes the most sense for developers, power users, and teams who are already comfortable with Ubuntu servers and want a private AI assistant. If you would prefer a managed setup, our AI Agent Development team builds and maintains production agent systems.

Prerequisites: Hardware, Software, and Accounts

Before running a single command, verify you meet these requirements.

Hardware minimum:

• NVIDIA GPU with at least 8 GB VRAM (RTX 3060 or better). GLM-4.7-Flash in 4-bit quantisation fits comfortably on 8 GB; the full GLM-4.7 model needs 16–24 GB.
• 16 GB system RAM (32 GB recommended for larger context windows).
• 50 GB free disk space for the OS, Ollama, and model weights.

Software:

• Ubuntu 22.04 LTS or 24.04 LTS (fresh install recommended).
• Node.js 20 or later (required by the OpenClaw CLI).
• An internet connection for the initial driver and model downloads.

Accounts:

• A Telegram account. You will create a bot via BotFather — no additional subscription is needed.
• Optional: an npm account if you want to publish custom OpenClaw skills.

Step 1 — Install Ubuntu and NVIDIA Drivers

If you are on a fresh Ubuntu install, the first task is getting your NVIDIA GPU recognised and the drivers stable. Skipping this step or using an out-of-date driver is the single most common cause of Ollama GPU acceleration failures.

Update the system first:

sudo apt update && sudo apt upgrade -y
sudo reboot

Install the recommended NVIDIA driver:

sudo ubuntu-drivers autoinstall
sudo reboot

After rebooting, confirm the GPU is visible:

nvidia-smi

You should see your GPU model, driver version, and VRAM usage. If nvidia-smi returns an error, run sudo ubuntu-drivers devices to list available drivers and install a specific version manually (e.g., sudo apt install nvidia-driver-550).

Install CUDA (optional but speeds up some model backends):

sudo apt install -y nvidia-cuda-toolkit
nvcc --version

Ollama’s built-in CUDA support does not require the full CUDA Toolkit — it ships its own CUDA libraries. Install the Toolkit only if you plan to run additional GPU workloads alongside Ollama.

Step 2 — Install Ollama and Pull GLM-4.7

Ollama is the local model server that OpenClaw talks to. It handles model downloads, quantisation formats, context window management, and the OpenAI-compatible REST API that OpenClaw’s Ollama provider uses.

Install Ollama:

curl -fsSL https://ollama.com/install.sh | sh

The install script detects your NVIDIA GPU automatically and configures the service. Verify Ollama is running:

ollama --version
systemctl status ollama

Pull the GLM-4.7-Flash model:

ollama pull glm-4.7-flash

This downloads approximately 5–6 GB of model weights. For the full GLM-4.7 model:

ollama pull glm-4.7

Confirm the model loaded correctly:

ollama run glm-4.7-flash "Say hello in one sentence."

If the model responds, Ollama is working. Exit with /bye.

Configure Ollama context window for OpenClaw:

OpenClaw’s agent layer benefits from a larger context window. Create or edit the Ollama model configuration:

ollama create glm-4.7-flash-ctx -f - <<EOF
FROM glm-4.7-flash
PARAMETER num_ctx 65536
EOF

Use glm-4.7-flash-ctx as your model name in the OpenClaw config (next step). A 64k context window lets OpenClaw maintain longer conversation histories and handle tool calls reliably.

Step 3 — Install OpenClaw and Run the Onboarding Wizard

With Ollama running, install the OpenClaw CLI via npm:

npm install -g openclaw

Verify the installation:

openclaw --version

If you have an old moltbot installation, the alias is preserved. You can run either openclaw or moltbot — they point to the same binary. To verify:

which moltbot   # should resolve to the same path as `which openclaw`

Run the onboarding wizard:

openclaw onboard

The wizard walks you through five setup stages:

1. Gateway initialisation — Creates ~/.openclaw/config.yaml and the local device key.
2. LLM provider selection — Choose “Ollama” from the list.
3. Ollama endpoint — Accept the default http://localhost:11434 or enter a remote Ollama host.
4. Model selection — Enter glm-4.7-flash-ctx (the custom context-extended model from Step 2).
5. Channel setup — Select Telegram. You will be prompted for a bot token (obtained in the next step).

Step 4 — Create Your Telegram Bot with BotFather

Telegram bots are managed through a special bot called BotFather. You need a bot token to link Telegram to OpenClaw.

1. Open Telegram and search for @BotFather.
2. Start a conversation and send /newbot.
3. BotFather asks for a display name (e.g., My OpenClaw Assistant) and a username ending in bot (e.g., myopenclaw_bot).
4. BotFather returns an API token in the format 123456789:ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefgh.
5. Copy this token — you will need it in the next step.

Step 5 — Connect Telegram to OpenClaw

Return to the openclaw onboard wizard. When prompted for your Telegram bot token, paste the token from BotFather.

The wizard configures your ~/.openclaw/config.yaml with a section like:

channels:
  telegram:
    enabled: true
    botToken: "123456789:ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefgh"
    dmPolicy: whitelist
    whitelist:
      - YOUR_TELEGRAM_USER_ID

To find your Telegram user ID, message @userinfobot in Telegram — it will reply with your numeric user ID.

Restart the OpenClaw gateway:

openclaw gateway restart

Check the gateway status:

openclaw status

You should see gateway: running, provider: ollama, and channels.telegram: connected.

Verify Everything Is Working

Open Telegram and send a message to your bot. You should receive a response within a few seconds (first response may take longer as the model warms up).

Test the full pipeline with a few prompts:

Troubleshooting Common Issues

Even with careful setup, a few failure modes appear repeatedly. Here are the fixes:

openclaw gateway fails to start:

• Check that Node.js 20+ is installed: node --version
• Verify Ollama is running: systemctl status ollama
• Check the gateway log: openclaw logs --tail 50

Telegram bot receives no response:

• Confirm the bot token in ~/.openclaw/config.yaml is correct.
• Check that your Telegram user ID is in the whitelist (if dmPolicy: whitelist).
• Test Ollama directly: curl http://localhost:11434/api/tags

Ollama runs on CPU instead of GPU:

• Run nvidia-smi — if the GPU is not listed, the driver is not loaded. Reboot and check.
• Run ollama logs to see if CUDA initialisation errors appear.
• Ensure you are not running Ollama inside a container without GPU passthrough.

High latency / slow responses:

• First response after model load always takes 5–15 seconds (warmup). Subsequent responses are faster.
• If all responses are slow, check GPU utilisation with nvidia-smi dmon. If GPU usage stays below 30%, the model may be exceeding VRAM and spilling to system RAM.
• Switch to glm-4.7-flash (not the full model) for faster inference on 8–12 GB cards.

For a deeper look at agent architectures and how to extend OpenClaw with custom tools, see our guide on building your own AI agent from scratch. If you want to connect OpenClaw to external data sources via the Model Context Protocol, read our MCP explainer.

Frequently Asked Questions

How Musketeers Tech Can Help

Setting up OpenClaw locally is a great starting point, but production-grade AI assistants for teams, customer support, or business automation require more than a single-server setup. Musketeers Tech designs and builds custom AI agent systems that integrate with your existing tools, scale reliably, and include monitoring and fallback logic that personal setups lack.

We have shipped AI assistant projects across diverse domains — from Chottay, an AI order-taking system for restaurants, to BidMate, an AI proposal assistant on Upwork. If you are considering moving from a personal OpenClaw instance to a team-wide or customer-facing deployment, our Generative AI Application Services team can architect the right approach — whether that means staying on local LLMs for privacy, using hybrid routing between local and cloud models, or integrating Model Context Protocol for tool connectivity.

Get Started Our AI Services View AI Portfolio

Final Thoughts

The OpenClaw setup on Ubuntu with NVIDIA, Ollama, and GLM-4.7 is the most practical path to a fully private, zero-cost AI assistant that lives in your Telegram. The five-step process — drivers, Ollama, GLM-4.7, OpenClaw onboarding, Telegram bot — takes under two hours on a clean Ubuntu install and produces a system that genuinely competes with cloud-based assistants on everyday tasks.

The MoltBot-to-OpenClaw rename is a cosmetic change on a maturing platform. If you are already running MoltBot, the upgrade is a single npm update -g openclaw and a gateway restart. If you are starting fresh in 2026, OpenClaw is the name to use.

For teams wanting to go further — adding retrieval-augmented generation, connecting to internal databases, or building multi-agent pipelines — read our deep-dive on autonomous workflow automation with AI agents or reach out to discuss a custom build.

AI-optimized version of this article: Read the text-only version