# Download the binary (Linux example)
curl -L https://github.com/localgpt-app/localgpt/releases/latest/download/localgpt-x86_64-unknown-linux-gnu.tar.gz | tar xz

# Run it
./localgpt --model llama3.2:latest --memory-path ./my_memory

# Your first command
/add-heartbeat "Check news headlines every 6 hours and summarize"

You just copied the exact commands to run a fully autonomous AI assistant that works completely offline. No API keys, no monthly fees, no data leaving your machine.

This isn't another ChatGPT wrapper. LocalGPT is a Rust reimagining of the OpenClaw pattern—built in just 4 nights—that gives you persistent memory in plain markdown files. Your AI remembers conversations from months ago because it writes them to disk, not some distant server.

TL;DR: Why This Changes Everything

• What: A 27MB Rust binary that runs a local AI assistant with persistent memory and autonomous tasks.
• Impact: Eliminates cloud dependency while maintaining semantic search and memory across sessions.
• For You: Complete privacy plus the ability to customize memory and skills without touching Python or Docker.

The Memory That Actually Works

Cloud AI assistants have goldfish memory. They forget context after a few messages unless you pay for expensive long-context windows.

LocalGPT solves this with three simple markdown files:

• MEMORY.md: Your conversation history
• HEARTBEAT.md: Scheduled autonomous tasks
• SOUL.md: Core personality and instructions

These files are human-readable and editable. Want to change how your AI behaves? Edit SOUL.md. Need to review what it remembered? Open MEMORY.md.

Search That Finds What You Mean

Most local AI tools have terrible search. You type "project ideas from last month" and get nothing.

LocalGPT combines two search methods:

• SQLite FTS5: Lightning-fast keyword search
• Local embeddings: Semantic understanding without API calls

This means you can search for "that coding discussion about Rust performance" and actually find it—even if you never used those exact words.

Autonomous Heartbeat Tasks

The killer feature? Your AI keeps working when you're offline.

Add a heartbeat task like "check my calendar every morning and remind me of meetings" and it just happens. No cron jobs, no scripts—just plain English instructions that persist across restarts.

Because everything's local, these tasks run instantly without waiting for cloud responses.

27MB vs. The Bloat

Compare LocalGPT's footprint:

• LocalGPT: 27MB single binary
• Typical Python setup: 500MB+ with dependencies
• Docker container: 1GB+ with base images
• Node.js equivalent: 200MB+ with node_modules

That's not just smaller—it's portable. Drop the binary on any machine and run it. No package managers, no environment setup, no compatibility issues.

Built for Hackers, Not Just Users

The OpenClaw compatibility is intentional. If you've used OpenClaw before, your existing memory files work immediately.

But the Rust implementation brings serious advantages:

• Memory safety: No segfaults from memory leaks
• Performance: Native speed for search and embeddings
• Reliability: Proper error handling instead of Python exceptions

This is a tool that won't break when you add 10,000 memories or schedule 50 heartbeat tasks.