Prerequisites
Before installing AutoGPT you need:
- OpenAI API key — platform.openai.com — GPT-4 access required
- Docker Desktop (recommended) — docker.com — v20+
- Git — for cloning the repo
- 8GB RAM minimum (16GB recommended)
The Docker method is recommended because AutoGPT executes code — Docker isolates it from your host system.
Method 1: Docker Install (Recommended)
Step 1: Clone the Repository
git clone https://github.com/Significant-Gravitas/AutoGPT.git
cd AutoGPT/classic/original_autogptStep 2: Configure Environment
cp .env.template .envOpen .env and set your keys:
OPENAI_API_KEY=sk-your-openai-key-here
# Optional but recommended
SMART_LLM=gpt-4o
FAST_LLM=gpt-4o-mini
# Memory backend (local JSON by default)
MEMORY_BACKEND=local
# Limit costs: stop after N API calls
# FAST_TOKEN_LIMIT=4000
# SMART_TOKEN_LIMIT=8000Step 3: Build and Run
docker compose run --rm auto-gpt --speakThe --speak flag enables audio output. Remove it for silent mode.
On first run, Docker will pull the base image (~2GB). Subsequent runs start in seconds.
Step 4: Set Your Goal
AutoGPT will prompt you:
Name your AI: MyResearcher
Your AI's role: An AI assistant that researches AI agent frameworks
Goal 1: Find the top 5 Python AI agent frameworks by GitHub stars
Goal 2: Write a comparison table in Markdown
Goal 3: TerminatePress Enter after each goal. Type nothing for Goal 4 to start.
AutoGPT will begin planning and executing autonomously.
Method 2: Local Python Install
Use this if you can’t run Docker.
Step 1: Requirements
# Python 3.10+ required
python --versionStep 2: Clone and Install
git clone https://github.com/Significant-Gravitas/AutoGPT.git
cd AutoGPT/classic/original_autogpt
python -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
pip install poetry
poetry installStep 3: Configure and Run
Same .env setup as Docker, then:
python -m autogptMonitoring Token Usage
AutoGPT can burn through API credits quickly. Enable budget limits:
# In .env
API_BUDGET=2.00 # Stop after spending $2.00Check spending in real time at platform.openai.com/usage.
For a typical goal (5 subtasks, web search), expect $0.10–$0.50 with GPT-4o-mini or $0.50–$2.00 with GPT-4o.
Workspace Files
All files AutoGPT creates are saved to:
AutoGPT/classic/original_autogpt/autogpt/workspace/You’ll find research notes, drafts, code files, and any other outputs here.
Common Issues
“ModuleNotFoundError: No module named ‘autogpt’”
You’re not in the right directory. Make sure you’re in AutoGPT/classic/original_autogpt/ before running.
“openai.AuthenticationError: Incorrect API key”
Check .env — the key should start with sk-, no quotes, no trailing spaces.
Agent loops indefinitely
Set CYCLES_LIMIT=10 in .env to stop after 10 reasoning cycles.
Frequently Asked Questions
Does AutoGPT work with Claude or Gemini instead of GPT-4?
The classic AutoGPT CLI is tightly coupled to OpenAI’s API. For Claude or Gemini, use the new AutoGPT Platform (web interface) which supports multiple providers, or consider CrewAI which natively supports any LiteLLM model.
How do I update AutoGPT to the latest version?
Pull the latest code and rebuild: git pull && docker compose build. Check the releases page for breaking changes before updating.
Can I run AutoGPT on a server without a GPU?
Yes. AutoGPT uses OpenAI’s API — all LLM computation happens on OpenAI’s servers. The local machine only needs CPU for orchestration logic and tool execution (web requests, file I/O). A $5/month VPS is sufficient.
Next Steps
- What Is AutoGPT — Understand the concepts behind what you just installed
- Introduction to LangChain — Build more controlled agents with LangChain