AutoGPT

1. What is AutoGPT?

AutoGPT gives you a "team of AI experts". You describe a job in plain English, a coordinator called Otto hands it to the right specialists (SEO content, prospect research, executive assistant and so on) and brings the results back to you. Next to the chat there is Build, a drag-and-drop editor for turning steps into a workflow that runs on demand or on a schedule, and a Marketplace of ready-made workflows you can adapt.

There are two ways to use it:

• The hosted platform (platform.agpt.co): nothing to install, but it is paid. Pro is 50 USD/month (7-day trial, card required), Max is 320 USD/month, and workflow runs are billed on top by usage. Prices can change, see https://agpt.co/pricing.
• Self-hosting on your own machine (this guide): no license fee, but you provide the computer, your own AI model keys (or a model running locally), and you handle updates and backups.

The latest release at the time of writing is AutoGPT Platform v0.11.0, published on 4 November 2026. The classic/ folder in the repository is the original 2023 experiment and is no longer supported; this guide does not cover it.

A good fit if:

• You want AI to handle recurring work: drafting articles, summaries, research, email prep.
• Your computer is reasonably powerful (8 GB RAM or more is recommended) and you are willing to install Docker.
• You want your data to stay on your own machine instead of a vendor's servers.

Not a good fit if:

• Your computer is low on RAM or disk space.
• You do not want to touch a command line at all. The paid hosted platform is easier then.
• You plan to resell the platform as a competing hosted service: the license does not allow it (see section 7).

2. What you need

• A Windows, macOS or Linux computer that can run Linux containers on amd64 or arm64 (almost any recent computer).
• Memory: at least 8 GB RAM is recommended. The developers' test installs used about 5–6 GB at startup and while running.
• Disk: about 25 GB free. SSD storage is recommended.
• Docker (software that runs applications in "containers", sealed boxes that hold everything the app needs). On Windows and macOS install Docker Desktop: https://docs.docker.com/desktop/ . On Linux install Docker Engine: https://docs.docker.com/engine/install/ . Docker Desktop has its own license terms from Docker.
• AI model keys for chatting with Otto and the experts. The default setup uses OpenRouter (https://openrouter.ai) for chat and OpenAI (https://platform.openai.com) for memory. Both charge by usage. Without keys you can still sign in and use Build; only model-backed features report missing credentials.
• Or Ollama (https://ollama.com) if you would rather run a model on your own computer with no API bill (this needs a stronger machine).

You do not need Git, Node.js, npm or Docker Compose.

3. Step-by-step installation

The easiest route is the official installer: one command downloads the latest release, checks it and starts AutoGPT in a single container.

Step 1: Install and start Docker

Install Docker from the links in section 2, then open Docker Desktop (Windows/macOS) and wait until it says it is running. Check it with:

docker info

What you should see: a long block of information and no connection error. If you get an error, see section 5.

Step 2: Run the installer

Windows (paste into PowerShell):

& {
  $installer = Join-Path ([IO.Path]::GetTempPath()) "autogpt-install-$([guid]::NewGuid().ToString('N')).ps1"
  try {
    curl.exe --proto '=https' --proto-redir '=https' --tlsv1.2 -fsSL -o $installer https://setup.agpt.co/install.ps1
    if ($LASTEXITCODE -eq 0) {
      powershell.exe -NoProfile -ExecutionPolicy Bypass -File $installer
    }
  } finally {
    Remove-Item -LiteralPath $installer -Force -ErrorAction SilentlyContinue
  }
}

macOS and Linux (paste into Terminal):

(
  installer="$(mktemp)" &&
  trap 'rm -f "$installer"' EXIT &&
  curl --proto '=https' --proto-redir '=https' --tlsv1.2 -fsSL \
    -o "$installer" https://setup.agpt.co/install.sh &&
  bash "$installer"
)

The installer first checks memory, disk and Docker, then downloads AutoGPT, creates a data volume called autogpt-platform-data and starts a container called autogpt. The first run can take several minutes because there is a lot to download.

What you should see: the installer reports a healthy container. Running it again with nothing changed does nothing extra, so a second run by mistake is harmless.

Note: To only check whether your computer qualifies, add --preflight-only (macOS/Linux, after bash "$installer") or -PreflightOnly (Windows, after -File $installer).

Step 3: Open AutoGPT and create your account

Open a browser at http://localhost:3000 and sign up. Passwords need at least 12 characters.

What you should see: the AutoGPT home screen with Home, Agents, Marketplace and Build along the top.

Step 4: Make your account the administrator

Replace you@example.com with the email you just signed up with, then run:

docker exec autogpt autogpt-admin promote you@example.com

Sign out and sign back in.

Step 5: Close sign-ups to strangers

Right after install, anyone who can reach the address can sign up. Open the autogpt.env settings file in the installer's private folder:

• Windows: %LOCALAPPDATA%\AutoGPT
• macOS/Linux: ~/.config/autogpt

Add the line:

AUTH_ALLOW_NEW_ACCOUNTS=false

Then remove the old container (your data is kept) and run the Step 2 command again:

docker stop --time 360 autogpt
docker rm autogpt

What you should see: the sign-up page no longer creates new accounts, and your own account still signs in.

Step 6: Add model keys or connect a local model

In the same autogpt.env file, pick one of the two options, save the file, then repeat "remove the container and rerun the installer" from Step 5.

Option A, API keys (replace each YOUR_... with your real key):

CHAT_USE_LOCAL=false
CHAT_USE_OPENROUTER=true
OPEN_ROUTER_API_KEY=YOUR_OPENROUTER_KEY
OPENAI_API_KEY=YOUR_OPENAI_KEY

Option B, a local model through Ollama. Install Ollama, set OLLAMA_CONTEXT_LENGTH=32768 for the Ollama server, and pull the two models the AutoGPT docs use as their example:

ollama pull hf.co/unsloth/Qwen3.5-4B-GGUF:Q4_K_M
ollama pull nomic-embed-text

Then add to autogpt.env:

CHAT_USE_LOCAL=true
CHAT_BASE_URL=http://host.docker.internal:11434/v1
CHAT_API_KEY=ollama
CHAT_FAST_STANDARD_MODEL=hf.co/unsloth/Qwen3.5-4B-GGUF:Q4_K_M
GRAPHITI_LLM_MODEL=hf.co/unsloth/Qwen3.5-4B-GGUF:Q4_K_M
GRAPHITI_RERANKER_MODEL=hf.co/unsloth/Qwen3.5-4B-GGUF:Q4_K_M

Note: Never share autogpt.env; it holds your API keys. Details and other cases (Anthropic directly, Linux with Docker Engine) are at https://agpt.co/docs/platform/self-hosting/single-container .

4. First use

1. Go to http://localhost:3000 and sign in.
2. On Home, type a concrete job into the chat, for example: "Summarise the three main points of the article at this link". Otto takes the job, may bring in a suitable expert, works step by step and returns the result in the chat; files it creates appear in the side panel.
3. To build your own process, click Build: drag blocks in, connect one block's output to the next block's input, press run and watch each block execute.
4. Not sure where to start? Open Marketplace, pick an existing workflow and adapt it to your job.
5. Open Agents to follow your runs.

Note: If you use paid API keys, set spending limits. Self-hosted installs have no limit by default; the single-container docs explain how to turn on daily and weekly caps.

5. Common problems and fixes

Symptom	Cause	Fix
docker is "not recognized" / "command not found", or docker info cannot connect	Docker is not installed or Docker Desktop is not running	Install Docker as in section 2, open Docker Desktop, wait for it to start, try again
The installer fails to download AutoGPT right after a new release	The release has not finished publishing to Docker Hub	Wait a little and run the installer again
The container stays "starting" or "unhealthy"	First boot takes several minutes, or a service inside failed	Wait longer; if it persists, read the log with docker logs --tail 100 autogpt
The container keeps restarting, out-of-memory messages	Docker does not have enough memory	Docker Desktop: Settings → Resources, raise memory above 6 GB. Windows with WSL 2: set memory= under [wsl2] in %UserProfile%\.wslconfig, then run wsl --shutdown
Sign-up says "Email Not Allowed"	Sign-ups are closed or the email is not on the allow list	Check the AUTH_ALLOW_NEW_ACCOUNTS line in autogpt.env
Chatting with Otto returns error 401	The API key is missing or wrong	Recheck Step 6, and remember to remove the container and rerun the installer after editing the file
A run stays RUNNING forever after a restart	Runs in progress when the container stopped are abandoned	Start a new run

6. Uninstall / update

Updating to a new release:

1. Back up first using the "Cold backup" section at https://agpt.co/docs/platform/self-hosting/single-container (the script is long, copy it from the source page).
2. Stop and remove the container (not the data):

docker stop --time 360 autogpt
docker rm autogpt

3. Run the Step 2 command again. To pin a specific version, add --release=vX.Y.Z (macOS/Linux) or -Release vX.Y.Z (Windows).

Uninstalling: the official docs have no dedicated uninstall section. Stop and remove the container as above. Your data (accounts, workflows, generated secrets) lives in the autogpt-platform-data volume; only when you are sure you no longer need it, delete it with Docker:

docker volume rm autogpt-platform-data

Note: Deleting this volume erases everything permanently.

7. Frequently asked questions

Does it cost money? Self-hosting has no license fee. You pay the model provider if you use paid API keys; a local model through Ollama has no API bill. The hosted platform is a paid service (Pro 50 USD/month, Max 320 USD/month at the time of writing).

Can my company use it? The platform code (autogpt_platform/) is under the Polyform Shield license: personal use and internal business use are allowed, reselling it as a hosted service that competes with AutoGPT is not. The rest of the repository is MIT-licensed.

Where does my data live? On your computer, in the Docker volume autogpt-platform-data. If you use API keys, whatever you send to a model goes through that model provider.

Can other people on my network use it? By default AutoGPT only listens on your own machine (127.0.0.1). To serve other machines, put it behind a reverse proxy with HTTPS, as described in the single-container docs.

Is the 2023 AutoGPT (Classic) still usable? The code is still in the classic/ folder but is no longer supported; treat it as reference material.