Creating a Project

Create a project in the OCI Generative AI service.

Create a project by using the Oracle Cloud Console. Configure data retention and memory settings for conversations and responses. To run code and work with files in isolated environments, also enable sandboxes for the project.

Tip

After you create a project, you can view its details and perform other tasks, such as updating it, moving it to a different compartment, managing its tags, or deleting it. Use the Actions menu (three dots) in the Console to access these tasks. For more information, see Listing a project.

To create a project, in the list page, select Create project. If you need help finding the list page, see Listing a project.

Basic Information

(Optional) Enter a name for the project. Start the name with a letter or underscore, followed by letters, numbers, hyphens, or underscores. The length can be 1 to 255 characters. If you don't enter a name, the system generates a name that you can change later.
The generated name has the format generativeaiproject<timestamp>. Example: generativeaiproject20260316042443
(Optional) Enter a description for the project.
Select a compartment to create the project in. The default compartment is the same as the list page, but you can select any compartment that you have permission to work in.

Data Retention

Configure how long responses and conversations can be stored. Response retention controls how long individual responses are preserved after generation before automatic deletion. Conversation retention controls how long conversations is stored after the most recent update. The maximum retention duration is 720 hours (30 days).

For Response retention (hours), enter how long to retain each response after generation.
For Conversation retention (hours), enter how long to retain a conversation after its most recent update.

Short-Term Memory Compaction

In Short-term memory compaction config, configure summarization of conversation context to preserve key information while reducing token usage.

Note

Compaction model can only be set during creation. It can't be changed later. Short-term memory compaction can be enabled at any time, but it can't be disabled or deleted once it has been enabled unless this project is deleted.

(Optional) Enable short‑term memory compaction to summarize earlier chat history and keep context lightweight.
If you enable short-term memory compaction, select a Compaction (summarization) model.
The list of models varies by region. For available models, see Generative AI Models and Regions for Enterprise AI Agents.

Long-Term Memory

In Long-term memory config, enable long-term memory to extract key information from conversations and store it for later retrieval. Stored information is embedded as searchable vectors.

Note

Extraction and embedding models can only be set during creation. They can't be changed later. Long-term memory can be enabled at any time, but it can't be disabled or deleted after it has been enabled unless this project is deleted.

Tip

If you enable this long-term memory feature, for best results we recommend that in the Data retention section, you set both the Response and the Conversation retention time to the maximum available value (720 hours).

(Optional) Enable long‑term memory to help store key information for retrieval in future conversations.
If you enable the long‑term memory feature, then select models:
- Long-term memory extraction model: The model for capturing and identifying key information to store. The list of models varies by region. For available models, see Generative AI Models and Regions for Enterprise AI Agents.
- Long-term memory embedding model: The model for converting the stored information into searchable vectors.

Sandbox Configuration

Enable sandboxes to create isolated runtimes for this project. Each sandbox can run commands and read or write files. The project supplies shared networking settings. See Sandboxes in OCI Generative AI and API-Level Permissions for Sandboxes.

Important

After you enable sandboxes for a project, you can't disable them. Select the networking configuration before creating sandboxes. To use a different custom subnet or set of network security groups, create another project.

(Optional) Expand Sandbox configuration, and turn on Enable sandboxes in this project.
New sandboxes created through the CLI or SDK use the project's sandbox configuration. You can leave sandboxes disabled if the project doesn't need an execution environment.
If you enable sandboxes, select a Networking mode.
- Managed: Use service-managed outbound networking. Select a Network policy mode that allows access to the destinations your workload needs. ALLOW_ALL allows outbound access. Review this selection before creating the project.
- Custom: Route outbound traffic through your subnet. Enter the Subnet OCID for a private subnet in the same region as the project. To add an optional network security group, select Add network security group tag and enter its Network security group OCID. You can provide up to five unique NSGs in the same VCN as the subnet.
With custom networking, your VCN routes, security lists, and NSGs control outbound access. See Networking for Sandboxes.
(Optional) In Warm pool configuration, configure capacity to prepare before sandbox requests arrive.
Warm pools are created after the project becomes active. Networking mode and warm pool type are separate choices: either networking mode can be used with either pool type.
- Managed (default): Use a service-managed runtime image. Enter a unique Name, select a Shape, and select the Desired idle-capacity target shown in the form.
- Custom: Use your own container image. Enter a unique Name and a Container image, then select a Shape and Desired idle-capacity target. Prepare the image and access permissions described in Using a Custom Image for a Sandbox.
Select Add pool to configure another pool. To create the project without warm pools, remove every pool entry by selecting its remove icon. See Warm Pools for Sandboxes.

Oracle Cloud Infrastructure Documentation