Summary
Create your first agent with Oracle AI Database Private Agent Factory from deployment in OCI up to a test that turns a PDF into a blog article. On this tour we connect the platform with Oracle Autonomous AI Database 26ai and we set up models using OCI Generative AI.
We have been preparing the Comparison, permissions, database user and connecting wallet. We then set up the VCN network and deploy the application from OCI Marketplace onto a virtual OCI Compute machine.
We continue with the creation of the initial user, the Autonomous connection test and the configuration of a generative model and an embeddings model. We have reviewed endpoint and API Key authentication from OCI to set up connections and save model settings.
Finally, we explore the templates and visual environment of Agent Builder. A from a template, we review the prompt, select the LLM and test a flow that generates a blog article from the content of a PDF. We then compare the result with the original document to review how you use the input information.
Start
Presentation of the Oracle AI Database Private Agent Factory tour: display the platform at OCI, connect with Autonomous and prepare a first agent that converts a PDF content into a blog article.
Documentation, sharing and policies
The walkthrough starts with the OCI Marketplace installation documentation and selects a compartment to group the example's resources. A policy is then created at the tenancy root to allow work within that compartment. The presenter explicitly describes this broad policy as a test configuration. This step prepares access for creating the database, network and application instance.
Creation of Oracle Autonomous AI Database
In Oracle Autonomous AI Database, the laboratory compartment is selected and instance creation begins. The walkthrough shows the workload options, selects version 26ai and adjusts compute capacity before setting the administrator password. After starting creation, provisioning must finish. The database needs to be available before preparing the user through Database Actions.
User and privilege in Database Actions
SQL is opened in Database Actions, and the script from the installation prerequisites is copied. The username and password placeholders are replaced with the values that Private Agent Factory will use. The statements are then run to create access and grant its privileges.
The demonstration checks the execution output: this editor rejects the EXIT instruction, while the preceding statements have executed. User creation and permission grants are checked before these credentials are used in the connection wizard.
Autonomous wallet download
The wallet is downloaded as a ZIP file from the Autonomous connection options, with a password to protect it. The presenter explains that this password can differ from the database administrator password. This file will later be used with the application database user and its password to test the connection from Private Agent Factory.
Creation of VCN network
Virtual Cloud Network is opened, the laboratory compartment is selected and the VCN creation wizard is used. The network is named and its provisioning is confirmed. Once available, its security configuration is reviewed. This VCN will later be selected when creating the Marketplace stack, placing the application instance in the prepared network.
Network rules and port of entry
The port rules used by the laboratory are added to the VCN security configuration. The walkthrough identifies port 8080 and the database connection port, and enters the corresponding network information. This preparation connects application access with connectivity to Autonomous. Before proceeding to Marketplace, the example's rules are checked in the network.
Deployment from OCI Marketplace
Oracle AI Database Private Agent Factory is located in the OCI Marketplace application catalog, and stack creation begins. The version, region, compartment, VCN and public subnet selected for the demonstration are reviewed. The virtual machine is also selected; the presenter mentions the E5 Flex and E6 options.
After applying the stack and waiting for deployment to finish, its output is checked. The resulting link opens the installation wizard in the browser, where the initial user and connections will be configured.
Initial user and connection with Autonomous
The wizard first creates the platform login user with an email address and password. The Autonomous connection is then configured: the wallet is uploaded, along with the previously prepared database user and its password. These are separate credentials, one for the platform and another for the database.
The connection test is run, and setup continues once it succeeds. During installation, the task log for the database and application is reviewed until the next configuration step becomes available.
LLM configuration in OCI Generative AI
Registration begins for the generative model that the platform will use. The chat is opened in the OCI Generative AI console, and a model from the Meta Llama family is selected. The endpoint comes from the console's code example, and the compartment identifier is copied from its OCI details page. These values prepare the provider configuration, which is then completed with authentication details.
API Key Authentication from OCI
An API Key is prepared from the OCI profile, and its key file is added to the model configuration. The form collects the user identifier, fingerprint, tenancy and region, along with the previously defined endpoint and compartment. These details are used to test the connection. The walkthrough checks the result before saving the connection for the generative model.
embeddings model configuration
An embeddings configuration is added separately from the generative model configuration. The example reuses the OCI Generative AI endpoint, compartment and authentication details: user, fingerprint, tenancy and API Key file. Another connection test is run, and the messages confirming registration are reviewed.
With both configurations prepared, installation is completed and the platform user signs in. The generative model and the embeddings model are then available for the following steps.
Templates, Agent Builder and model management
After signing in, the walkthrough explores the main Private Agent Factory views: templates for different tasks, the knowledge base, Agent Builder and model management. Templates serve as references for creating other projects, while the visual editor organizes components and connections.
The demonstration also shows that additional LLM can be registered using the access details configured during installation. This connects infrastructure preparation with model selection and the design of the agent to be tested next.
Record another model LLM
An additional LLM is registered through model management, entering the access details requested by the platform again. The example leaves more than one model available when configuring an agent. The walkthrough then returns to the visual editor, where components are connected to define the flow and its tools.
Template to convert a PDF to blog
A template is selected to turn the contents of a PDF into a blog article. In Agent Builder, the prompt defining this transformation is reviewed, the LLM is selected and the input document prepared for the example is used. After saving the configuration, the flow is ready to run in the test environment.
Execution of agent and review of result
A test input is submitted to run the agent with the task defined in the prompt. The output takes the form of a blog article, following the selected template. The original PDF is then opened and compared with the generated text. This review shows how the flow uses the document's information to carry out the requested transformation.
Closing
The presenter closes the demonstration and previews future videos exploring the tool and adding more components. The walkthrough ends with the template agent executed and its result reviewed.