An Oracle cloud engineer has published OCI Atlas, a community plugin that teaches AI assistants how Oracle Cloud Infrastructure (OCI) services fit together. Oracle described it in a blog post on November 11, 2026. It works with ChatGPT and Codex where the environment allows it.
The project is released under the UPL 1.0 licence. It is not a managed Oracle service, and the author calls it a community project that welcomes contributions and corrections.
What OCI Atlas Contains
OCI Atlas packages OCI guidance and multi-service workflows as a plugin. An entry skill points the assistant to the right reference as a conversation develops, so it does not have to load everything at once.
Version 1 ships 73 references. Of these, 48 are service guides and 12 are cross-service recipes. Topics range from networks, compute and storage to databases, AI and ML, app platforms and day-to-day operations.
The recipes join single services into common jobs. Examples are private application deployment, data ingestion, serving a predictive model and connecting to a database.
Knowledge and Execution Stay Separate
The design splits three roles. The model reads the request, OCI Atlas supplies OCI knowledge, and the host's own tools reach the environment. That means the same plugin can support a pure design chat or a session where an assistant carries out steps.
Any resource action runs with your own signed-in OCI access and permissions. Which install and run routes you can use depends on your client and workspace.
Use Cases Oracle Describes
- Deployment planning: weighing Container Instances, OKE and Functions for a stateless containerized API, and listing the networking, permission and availability choices to settle first.
- Inventory: with an authenticated OCI CLI, listing visible compartments and running compute instances while staying inside the chosen tenancy, region and scope.
- Dependency design: a compute app that reaches Autonomous Database privately needs DNS, routing, security rules, TLS and database privileges. Object Storage through a service gateway is a separate path with its own authorization.
- Troubleshooting: start from the failing operation, the exact error, the target resource and where it ran. For a private database timeout, the checks run from name resolution and routes to the listener, TLS and finally SQL permissions.
The guidance also covers judgment calls, for example checking what really exists after a request times out. For existing environments, it considers whether Terraform or Resource Manager owns the configuration.
What It Means for Developers
Generic AI assistants often give plausible but wrong cloud advice, because they miss service-specific details such as which gateway carries which traffic. A curated reference set is one way to reduce that, and OCI Atlas shows the pattern for OCI.
Because it is open source, you can read the guides, correct them and add your own recipes. That matters if your team has its own naming, network or security standards.
Take care with any tool that can change infrastructure. Start with read-only tasks such as comparisons and inventories. Give the CLI profile the least privilege that works, and keep Terraform or Resource Manager as the source of truth for resources they manage.
How to Get Started
The OCI Atlas repository holds install steps, sample prompts, service guides and architecture walkthroughs. Oracle suggests beginning with a service comparison, an application design or a scoped resource inspection, then building a workflow around your own requirements.
Conclusion
OCI Atlas is a small but practical attempt to put OCI knowledge inside AI assistants. Try it on a design question or a read-only inventory first, and review every action it proposes before it touches a real tenancy.
Sources
See my articles more often when you search Google.
