Local LLM + OpenClaw

Your private AI agent that remembers conversations without leaking data.

Run AI models locally with LM Studio and give them real capabilities with OpenClaw or Hermes Agent. Your AI can control your mouse, execute commands, access files, and automate tasks — all without sending a single byte to the cloud.

What You're Setting Up

• LM Studio (v0.4.20) - Desktop app that runs AI models locally and serves them over an OpenAI-compatible API
• OpenClaw (v2026.7.x) - AI agent framework that can automate tasks on your computer using local models
• Hermes Agent (v0.19.0) - Alternative local-first agent runner by Nous Research, with self-improving skills

Why This Setup?

You must've heard about OpenClaw, instead of using it with a company owned LLM on cloud, consider a more privacy focused option.

🔒 Complete privacy — Everything runs on your machine

🧠 Persistent memory — Your AI remembers past conversations

⚡ Real actions — Not just chat, but actual task execution

🛡️ Security audits — Built-in tools to harden your setup

Quick Start

1. Get LM Studio Running

Windows:

irm https://lmstudio.ai/install.ps1 | iex

Or download from https://lmstudio.ai/download and run the installer.

Linux:

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

This installs the lms CLI and the llmster headless inference engine. For the GUI, download the AppImage from https://lmstudio.ai/download.

Mac:

curl -L -o lm-studio.dmg "https://lmstudio.ai/download/macos"
open lm-studio.dmg

Start the server:

lms server start --port 1234 --cors

Or use the GUI: go to Developer/Server tab → Start Server.

LM Studio serves models on localhost:1234.

2. Get OpenClaw

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

The onboard command walks you through setup interactively — pick LM Studio as your provider, enter http://127.0.0.1:1234/v1 as the base URL, and select your loaded model.

3. Start Everything

Make sure LM Studio's Server is running, then:

openclaw gateway --port 18789

Open http://localhost:18789. Done.

4. Harden Your Setup

openclaw security audit
openclaw security audit --fix

This checks for known vulnerabilities and applies recommended fixes.

Recommended Local Models (July 2026)

Model	Size	Best For	VRAM Needed
Qwen 3.x (27B)	27B	Daily driver, coding, agentic work	16-24 GB
Gemma 4 (12B)	12B	Great generalist, creative writing	8-16 GB
Phi-4 Mini (3.8B)	3.8B	Low memory, fast responses	4 GB
DeepSeek R1	1.5-32B	Chain-of-thought reasoning	8-24 GB
Llama 4 Scout	Various	Massive context window (10M tokens)	16+ GB

Alternatives Worth Knowing

Tool	What It Does	Role
Ollama	Headless CLI model server	LM Studio alternative (no GUI)
Jan.ai	Privacy-first desktop chat UI	LM Studio alternative (with GUI)
Open Interpreter	Terminal code execution agent	Interactive coding agent
Aider	Git-integrated coding agent	Pair programming tool
AnythingLLM	Desktop RAG & knowledge base	Document QA & chat

All of these can use the same local LM Studio or Ollama backend.

Full Guides

• LM Studio Setup
• OpenClaw Setup
• Hermes Agent Setup
• Connect to Telegram
• Connect to Discord
• Connect to WhatsApp
• Security Guide

⚠️ Security Notice

OpenClaw runs with high system privileges (shell, filesystem, credentials). Several critical CVEs were disclosed in 2026. Always:

1. Run openclaw security audit --fix after installation
2. Bind the gateway to 127.0.0.1 only
3. Use strong auth tokens
4. Run in a container or VM if possible
5. Keep OpenClaw updated

See Security Guide for full details.

Done. Enjoy your private AI. 🔒