Most AutoGPT tutorials cover the old version. The repo now has two products, and you’re probably chasing the wrong one.
This guide installs AutoGPT Platform v0.6.53 (March 25, 2026 release) – the new visual agent builder, not the 2023 command-line tool. Running locally in 20 minutes.
What You’re Actually Installing
The AutoGPT repository contains two products:
- AutoGPT Platform (new): Visual workflow builder. Drag blocks, connect them, create AI agents. Runs on Docker with a web UI at localhost:3000. This is it.
- AutoGPT Classic (original): Command-line autonomous agent from 2023. Still there, but not the focus.
Think of Classic as a self-driving car prototype that sometimes crashes into trees. Platform is the production model with lane assist and a steering wheel you can grab.
System Requirements That Actually Matter
The official GitHub README lists minimums. They’re real:
| Component | Minimum | Recommended |
|---|---|---|
| CPU | 2 cores | 4+ cores |
| RAM | 8GB | 16GB |
| Storage | 10GB free | 20GB+ free |
| OS | Ubuntu 20.04+ / macOS 10.15+ / Windows 10/11 with WSL 2 | Same |
8GB RAM? Docker containers struggle during startup if you’re running anything else. 16GB is where it stops complaining.
Windows: use WSL 2, not Hyper-V. Official docs are explicit – Hyper-V causes the Supabase database container to fail health checks. You’ll spend an hour troubleshooting before finding that buried in the setup guide.
Step 1: Install Prerequisites
Four things before cloning:
- Docker Desktop (20.10.0+): Download here. Make sure it’s running.
- Docker Compose (2.0.0+): Comes with Docker Desktop on Windows/Mac. Linux users install separately.
- Git (2.30+):
git --versionto check. Missing? Get it from git-scm.com. - Node.js (16.x+) and npm (8.x+):
node --version && npm --versionto verify. Install from nodejs.org.
Optional: VSCode. Repo includes dev container configs, official tutorial assumes you’re using it.
Watch out: On Windows, after installing Docker Desktop, open Settings → General and confirm “Use the WSL 2 based engine” is checked. If you already had Docker with Hyper-V, switching to WSL 2 erases existing containers – back them up first.
Step 2: Clone and Navigate
Terminal:
git clone https://github.com/Significant-Gravitas/AutoGPT.git
cd AutoGPT
Pulls the entire repo (Platform + Classic). You’ll work in autogpt_platform.
Step 3: Run the Setup Script (Or Don’t)
Official docs suggest ./run setup to auto-install. Works if Docker’s already running and your system meets requirements. Fails silently if Docker isn’t up.
First install? Skip the auto script. Do it manually – you’ll know what broke.
Manual Setup
Go to the platform directory:
cd autogpt_platform
Copy the default environment config:
cp .env.default .env
.env contains database credentials, API keys, service ports. Defaults work for local development. Open it once to see what’s there. You’ll add LLM API keys later.
Step 4: Launch Docker Services
From autogpt_platform:
docker compose up -d --build
This builds frontend/backend Docker images, starts PostgreSQL/Redis/RabbitMQ/Supabase containers, runs database migrations. Takes 5-10 minutes on decent connection.
Watch for errors. supabase-db marked “unhealthy”? Stop everything (docker compose down), confirm Docker uses WSL 2 (Windows) or is properly configured (Linux/Mac), try again.
Step 5: Verify It’s Running
Browser → http://localhost:3000. You should see the AutoGPT Platform login screen.
Page doesn’t load?
- Run
docker ps– check all containers are up. Look forautogpt_platform-frontend-1,autogpt_platform-backend-1. - Logs:
docker compose logs frontendordocker compose logs backend - Port 3000 in use?
lsof -i :3000(Mac/Linux) ornetstat -ano | findstr :3000(Windows)
Backend API health: curl http://localhost:8006/health should return 200 OK.
Step 6: Add Your LLM API Keys
AutoGPT Platform doesn’t include AI models – calls external APIs. You need at least one:
- OpenAI: Key at platform.openai.com/account/api-keys. Paid account with billing enabled required.
- Anthropic (Claude): Keys from console.anthropic.com
- Others: Google DeepMind, Mistral, xAI, DeepSeek supported as of April 2026.
Open autogpt_platform/backend/.env, add your key:
OPENAI_API_KEY=sk-proj-xxxxxxxxxxxxxxxxxxxxxxxx
Restart backend:
docker compose restart backend
Common Install Errors (And Fixes)
1. “Bad git executable”
ImportError: Bad git executable
Git isn’t installed or not in system PATH. Install Git, restart terminal. Windows: make sure “Git from the command line and also from 3rd-party software” was selected during install.
2. “No module named autogpt”
Wrong directory. AutoGPT Classic runs from AutoGPT/classic/original_autogpt. AutoGPT Platform runs in Docker – this error means you’re trying Classic commands.
Want Platform? Ignore this, use the web UI. Want Classic? cd AutoGPT/classic/original_autogpt first.
3. “supabase-db unhealthy”
Docker Compose shows database container unhealthy, won’t start platform.
Windows: Docker’s using Hyper-V instead of WSL 2. Open Docker Desktop → Settings → General → Enable “Use the WSL 2 based engine” → Restart Docker. Then docker compose down && docker compose up -d --build.
4. Port 3000 already in use
Find what’s using it. Kill that process or edit autogpt_platform/.env to change frontend port.
5. Out of memory during build
You’re on the 8GB minimum. Docker doesn’t have enough RAM allocated. Docker Desktop → Settings → Resources → increase Memory to 6GB minimum (8GB better). Restart Docker.
Next: Create Your First Agent
AutoGPT’s running. Now:
- Log in at
localhost:3000 - Click “Create Agent” in Agent Builder
- Drag blocks from left panel – Trigger → LLM Prompt → Output
- Configure each block (select LLM provider, write prompts)
- Click “Run” to test
Official docs have examples: Reddit-to-video agents, YouTube summarizers. Start simple – trigger on schedule, call API, save result to file.
Upgrading and Cleanup
Update to newer version:
cd AutoGPT
git pull origin main
cd autogpt_platform
docker compose down
docker compose up -d --build
Full uninstall:
docker compose down -v # -v removes volumes (database data)
cd ..
rm -rf AutoGPT
API keys live in .env – back that up if reinstalling later.
FAQ
Do I need GPT-4 access or can I use GPT-3.5?
GPT-3.5 works and costs less. But GPT-4 is better for agents – Classic got stuck in loops with GPT-3.5 because the model made mistakes, then the agent tried to fix them by making more mistakes. Platform handles this better but still benefits from GPT-4’s reasoning.
Is AutoGPT Platform free?
Software’s open-source, free to self-host. You pay for LLM API calls (OpenAI, Anthropic, etc.). One agent run: few cents to several dollars depending on steps taken and model used. My first agent? $0.40 for a simple task. Then I misconfigured a loop and burned $12 before I noticed. Watch API usage at first.
What’s the difference between AutoGPT Platform and Classic?
Classic (2023): command-line tool. You give it a text goal, it autonomously breaks into sub-tasks, executes them, tries to achieve the goal without human input. Famous for being ambitious but unreliable – stuck in loops, hallucinated URLs, racked up API bills. Platform: complete rebuild. Visual workflow builder where you design agent logic as a graph of blocks (triggers, LLM calls, integrations, logic). More control, avoids some of Classic’s chaos. But less “autonomous” in the original sense – you’re building the logic, not just stating a goal. Want the viral 2023 experience? Classic. Want something that works? Platform. For what it’s worth, the repo now has both under different licenses – Platform’s under Polyform Shield (non-commercial restrictions), Classic’s still MIT as of April 2026.
Install the latest AutoGPT Platform version – docker compose up in the autogpt_platform directory, add API keys, open localhost:3000. Done.