PentAGI

1. What is PentAGI?

PentAGI is an open-source system that uses AI to assist with penetration testing ("pentesting" for short): probing a system for security weaknesses so they can be fixed before an attacker finds them. It drives familiar security tools (such as nmap, sqlmap and metasploit) inside an isolated Docker environment, and it has a web interface for managing the work.

Important legal warning, read before installing: PentAGI may only be used to test systems, servers and networks that you own or have explicit written permission from the owner to test. Scanning or attacking someone else's system without permission is against the law in most countries. The project's end user license agreement (EULA) states that it is for lawful, authorized use only. This guide stops at installation and a first test run.

Plainly: this is a specialist tool for security professionals, not a general-purpose app. It needs Docker, a command line and at least one AI model API key (which usually costs money).

A good fit if:

• You work in network security or are learning about it, and you have a test environment of your own (a virtual machine, a lab).
• You are comfortable with Docker and the command line.

Not a fit if:

• You have no system you are authorized to test.
• You want a click-and-go app and do not want to install Docker or set up API keys.

2. What you need

• A computer running Linux, macOS or Windows (the installer has builds for all three). The project's main requirement is Docker.
• Minimum specification: 2 or more CPU cores, 4 GB of RAM and 20 GB of free disk. You need an internet connection to download images and to call the AI model.
• Docker and Docker Compose (or Podman): software that runs the components in isolated "boxes". Download from https://www.docker.com/
• At least one AI model API key (required; PentAGI does not run without it): for example a key from OpenAI, Anthropic or Google Gemini. These services usually charge by usage. (A local model through Ollama is possible if your machine is powerful enough.)
• An isolated environment: it is wise to run PentAGI in a virtual machine or on a separate network for safety.

3. Step-by-step installation

There are two routes: the interactive installer (recommended) or a manual build with Docker Compose.

Route A, Step 1: Download the interactive installer

The installer is a text-based interface that runs in a command window. It checks your machine, writes the configuration file, helps you enter your API keys and starts the system. Download the build for your system from the official links, for example for Linux (64-bit Intel):

mkdir -p pentagi && cd pentagi
wget -O installer.zip https://pentagi.com/downloads/linux/amd64/installer-latest.zip
unzip installer.zip

The other builds, all named installer-latest.zip, are: Linux arm64 https://pentagi.com/downloads/linux/arm64/installer-latest.zip, Windows https://pentagi.com/downloads/windows/amd64/installer-latest.zip, macOS Intel https://pentagi.com/downloads/darwin/amd64/installer-latest.zip and Apple Silicon macOS https://pentagi.com/downloads/darwin/arm64/installer-latest.zip. Download only from these official links (they are listed in the README at https://github.com/vxcontrol/pentagi).

Route A, Step 2: Run the installer

On Linux the installer talks to Docker, so it needs suitable privileges. The README's recommended way is to run it as root:

sudo ./installer

(On Windows and macOS, run the installer program from the unzipped archive.) Follow the on-screen steps: system check, entering your AI model API key, setting passwords, then letting the installer start PentAGI.

What success looks like: the installer reports its checks as passed (Docker, network), then finishes starting the containers.

Route B, Step 1: Build it yourself (download the sample configuration)

If you know Docker you can set it up by hand. Create a folder and fetch the sample configuration plus the two sample provider files that go with it:

mkdir pentagi && cd pentagi
curl -o .env https://raw.githubusercontent.com/vxcontrol/pentagi/master/.env.example
curl -o example.custom.provider.yml https://raw.githubusercontent.com/vxcontrol/pentagi/master/examples/configs/custom-openai.provider.yml
curl -o example.ollama.provider.yml https://raw.githubusercontent.com/vxcontrol/pentagi/master/examples/configs/ollama-llama318b.provider.yml

Route B, Step 2: Fill in your API keys

Open the .env file in a text editor and fill in at least one AI model API key, for example:

OPEN_AI_KEY=your_openai_key
ANTHROPIC_API_KEY=your_anthropic_key
GEMINI_API_KEY=your_gemini_key

Also change the default passwords and secrets (such as COOKIE_SIGNING_SALT and the database passwords) to random strings of your own before real use.

PentAGI also uses an "embedding" model (it turns text into numbers for memory and search). By default this is OpenAI: if you leave EMBEDDING_URL and EMBEDDING_KEY empty, it uses OPEN_AI_KEY. If you only have a key from another company, see the embedding section of the official README to configure it separately.

Route B, Step 3: Download the startup file and run

curl -O https://raw.githubusercontent.com/vxcontrol/pentagi/master/docker-compose.yml
docker compose up -d

Watch progress with docker compose logs -f.

What success looks like: the containers download and move to the running state without startup errors.

4. First use

Step 1: Open the interface and sign in

Open your browser at https://localhost:8443. Because PentAGI uses a self-signed certificate, the browser will warn that the connection is not secure; on your own machine, choose to continue. The default account is admin@pentagi.com with the password admin.

Note: change the default password right after your first sign-in, before using the instance for anything real.

Step 2: Create a test flow

Open Flows in the sidebar and click New Flow. Choose a mode:

• Assistant: the AI proposes steps interactively and you steer it. Choose this for your first run so you can watch what happens.
• Automation: the AI carries out the whole task toward the goal on its own.

Pick the AI model provider you configured.

Step 3: Declare the target and the permitted scope

Describe the goal in plain language, and enter only addresses and systems that you are allowed to test (for example a virtual machine in your own lab). State the scope and limits clearly. In Assistant mode, watch each step the AI proposes.

What success looks like: PentAGI opens the flow page, showing the steps the AI plans to take and a log of tool activity. When the flow has produced results, the Report menu on the flow page lets you view the report, copy it, or download it as Markdown or PDF.

5. Common problems and fixes

Symptom	Cause	Fix
The browser warns the certificate is not secure at https://localhost:8443	PentAGI uses a self-signed certificate	On your own machine this is normal; choose to continue
docker: command not found or the containers will not start	Docker is not installed or not running	Install Docker from https://www.docker.com/ and make sure it is running before you type the commands
Pulling Docker images fails (restricted network)	The machine cannot reach Docker Hub	Make sure it can reach docker.io; if you use a proxy, configure it for Docker itself, not just PentAGI. See the official Docker documentation
A flow was created but sits there making no progress	Commonly a wrong or unreachable embedding (vector) model configuration	Read the log with docker logs pentagi and look for 401/403/404 errors or timeouts; re-check the API key and model names in .env; test the embedding part with the built-in tool docker exec -it pentagi /opt/pentagi/bin/etester test -verbose
Error "failed to select primary docker image via llm call"	The problem is with the configured AI model, not Docker (the flow's first model call chooses the Docker image, so this usually means the model server did not answer validly)	Check that the API key and model you use are valid
macOS says the installer is malware	The Mac blocks downloaded software	Download only from the project's official links, pick the build for your chip, then go to System Settings, Privacy & Security and choose "Open Anyway" if you trust the source

6. Uninstalling and updating

• Stop the system: in the install folder run docker compose down.
• Update: if you used the installer, open it again and choose Maintenance, Update PentAGI (this entry appears only when the installed version is not known to be the newest). If you built it by hand, download the newest docker-compose.yml, run docker compose pull, then docker compose up -d.
• Remove everything: after docker compose down, delete the install folder. Note that each flow's data sits in a data folder and may have to be deleted by hand to reclaim the space.

7. Frequently asked questions

Does it cost anything? PentAGI itself is free and open source (MIT license). But it calls AI models through API keys, and AI services usually charge by usage. With a local model (Ollama) there is no AI charge, but you need a powerful machine.

Does it need the internet? Yes, to download Docker images and call the AI model (unless you use a local model).

May I scan other people's websites? No. Only test systems you own or are authorized in writing to test. This is a binding requirement of the project's terms of use and also the law.

Is it safe for my computer? Run it in a virtual machine or an isolated network, because the tool contains real exploitation components. Do not run it on your main work machine.

What is the latest version? At the time of writing, the release marked latest is v2.1.0 (29 May 2026), but the source code is updated continuously. See the official pages: https://github.com/vxcontrol/pentagi and https://pentagi.com.