Oracle AI Database Private Agent Factory is a self hosted platform for building AI agents that run entirely on your own infrastructure. Nothing leaves your machine: not your data, not your prompts, not your documents. In this post I walk through a complete installation on a MacBook with Apple Silicon, from download to a running and configured web application, using the exact commands and outputs from my own install. The process has two halves: first the terminal installation, which builds and deploys the containers, and then the in application setup wizard, where you create your admin user, install the components, and connect a language model.
What you need before starting
A Mac with an M series chip, at least 16 GB of RAM, around 30 GB of free disk space, and Homebrew installed. The installer handles everything else, including Podman and the container images.
One tip before you begin: the installation creates a Podman virtual machine, and its default memory allocation is too small for the database container. After the machine is created, give it at least 8 GB:
podman machine stop
podman machine set --memory 8192
podman machine start
Part 1: The terminal installation
Step 1: Download the correct kit
Go to the official download page:
https://www.oracle.com/database/technologies/private-agent-factory-downloads.html
There are two installer kits and picking the right one matters more than anything else in this guide. Check your architecture first:
uname -aIf the output ends with arm64, as mine does on this MacBook, download the ARM 64 bit kit, in my case oracle_agent_factory_arm64_26.4.1.tar.gz. The X86 kit is only for Intel based Macs. The kits contain Linux container images, which is why the platform labels say Linux; everything runs inside a Podman virtual machine on your Mac, and the label refers to the CPU architecture, not the operating system.
Step 2: Extract the kit into a staging directory
mkdir ~/afo && cd ~/afo
cp ~/Downloads/oracle_agent_factory_arm64_26.4.1.tar.gz .
tar xzf oracle_agent_factory_arm64_26.4.1.tar.gz
This directory becomes the home of your installation. All future commands run from here.
Step 3: Run the interactive installer
bash interactive_install.sh --reset
bash interactive_install.sh
The reset flag clears any previous installer state so every step runs fresh. Answer N to the proxy question unless you are behind a corporate proxy. The installer then handles the prerequisites automatically: it installs bash, podman compose and wget through Homebrew, configures Podman with libkrun for Apple Silicon, and starts the Podman machine.
[INFO] Detected architecture: arm64
[INFO] Configuring Podman with libkrun for Apple Silicon...
[SUCCESS] Podman configured with libkrun for Apple Silicon.
[SUCCESS] Step 'Install Homebrew dependencies' completed.
[SUCCESS] Step 'Initialize and start Podman machine' completed.
When the installer asks about manual database setup, answer N. Quickstart mode will create and configure a local Oracle Database for you, which is the easiest path for a first installation. Manual setup is only for connecting to an existing Oracle AI Database instance.
Step 4: Build the container images
The installer runs make build and asks for the installation mode. Choose quickstart and confirm. The build pulls the base images from Oracle’s container registry and assembles the application image. Expect this to take a while on the first run since the database image alone is 8.5 GB. When it finishes you get a summary table, and both rows must say SUCCESS before continuing:
| Image | Status | Size | Tags |
| applied-ai-label | SUCCESS | 6.4 GB | localhost/applied-ai-label:26.4.0.0.0 |
| db23aifree | SUCCESS | 8.5 GB | .../database/free:23.7.0.0 |
Step 5: Deploy the containers
The installer moves straight into make install and asks the mode question once more. Choose quickstart again and confirm. Two prompts follow. First, a database admin password for the one time local database bootstrap: pick a password with at least 8 characters including upper case, lower case and a digit, and write it down. Second, the web UI access mode: for a laptop installation, choose option 1, private to this host.
The deployment then starts both containers and waits for the database to initialize. The first bootstrap creates an entire database from scratch, so be patient with the health check loop. Success looks like this:
| Stage | Status |
| Stopping Oracle Private AI Agent agent_factory | Successful |
| Storing App Secret Key | Successful |
| Database Migration | Successful |
| Starting Oracle Private AI Agent agent_factory | Successful |
| Configured Version Link | Successful |
The Oracle AI Database Private Agent Factory webapp can be accessed at:
https://localhost:8080/agentFactory/
✓ Quickstart deployment complete.
[SUCCESS] Installation script finished!
The terminal half is done. Everything else happens in the browser.
Part 2: Configuration inside the application
Step 6: Open the application and create your admin user
Open the address the installer printed:
https://localhost:8080/agentFactory
Note the https. Your browser will show a certificate warning because the application uses a self signed certificate; that is expected for a local installation, so click through it. You land in a setup wizard with three stages shown across the top: user configuration, installation, and LLM configuration. The first screen is the sign up form for the administrator account, the user responsible for managing the application and its other users. Enter an email and password.
Step 7: Install the components
The second stage installs Agent Factory into the configured database. Click Install and watch the installation log: it registers your user, configures the read only database credentials, registers the default data source, and installs and starts each component in turn.
The log runs through the Knowledge Agent, the Deep Data Research agent and the other components. Each block should end with a success line. When the log tells you to proceed to the next step for LLM Configuration, the application restarts briefly and the Next button becomes enabled.
Step 8: Set up a local LLM with Ollama
The third stage connects a language model, and the beauty of a private installation is that the model can be local too. Before filling in the form, install Ollama on your Mac and pull a model:
brew install ollama
brew services start ollama
ollama pull llama3.1:8bThe model is about 4.9 GB. Once the pull reports success, allow connections from the Podman virtual machine, since Ollama listens only on localhost by default:
launchctl setenv OLLAMA_HOST "0.0.0.0"
brew services restart ollama
Back in the wizard, the LLM Configuration page shows the generative model form with placeholder values:
Select the Ollama provider and fill in your real values: Model ID llama3.1:8b, URL host.containers.internal, Port 11434. The URL is important. From inside the container, localhost means the container itself; host.containers.internal is Podman’s alias for your Mac, where Ollama actually listens. Click Test connection, and once the Connection successful toast appears, save the configuration
Step 9: Configure the embedding model and finish
Below the generative model comes the embedding model, which converts your documents into vectors for search. Keep the Local provider, pick the bundled multilingual e5 base model from the dropdown, test the connection, save, and click Finish Installation.
Step 10: Verify the configuration
After finishing the wizard you land in the application itself. Under Settings and then Model Management you can confirm the result: the generative model configuration is listed with the Ollama provider and an Active status. From here the full application is yours: Knowledge Agents, Agent Builder, Prompt Lab and the rest of the sidebar.
Day to day operation
After a reboot, bring everything back with:
podman machine start
podman start oracle-database oracle-applied-ai-label
Give the database a minute to warm up, then open https://localhost:8080/agentFactory/ (I had to change to the current name of my server instead of the local host or the IP) again and log in with your admin account.
Lessons from my installation
The single most important thing in this whole guide is downloading the kit that matches your CPU architecture. My first attempt used the X86 kit on an Apple Silicon Mac, and every container crashed with cryptic qemu errors until I understood why. Run uname -a before you download anything.
The second is memory. The Podman machine defaults to 2 GB, and the database container alone needs more than that. Set 8 GB before deploying and the health checks pass without drama.
And the third is patience with the first database bootstrap. It genuinely builds a full Oracle Database inside a VM on your laptop, and the waiting loop is normal, not a hang.
With the platform running and a local llama connected, you have a complete private AI stack on a laptop: agent builder, vector capable database, and language model, with no cloud dependency and no per request cost. In my next post I use it to build a meeting booking agent that triages emails and drafts replies.
*The views expressed here are my own and do not represent those of my employer.*
Hello, I’m Bruno — a dual citizen of Brazil and Sweden. I bring a global perspective shaped by experiences in both South America and Europe, with a strong focus on collaboration and innovation across cultures. I am a Computer Scientist, PhD Candidate in Information and Communication Technologies, focusing on Data Science and Artificial Intelligence, and hold dual Master’s degrees in Data Science and Cybersecurity. With over fifteen years of international experience spanning Brazil, Hungary, and Sweden, I have collaborated with global organizations such as IBM, Playtech, and Oracle, as well as contributed remotely to projects across multiple regions. My professional interests include Databases, Cybersecurity, Cloud Computing, Data Science, Data Engineering, Big Data, Artificial Intelligence, Programming, and Software Engineering, all driven by a deep passion for transforming data into strategic business value.