docs(llm): document OCI Generative AI setup

openhands/usage/agent-canvas/model-configuration.mdx

@@ -49,6 +49,51 @@ Use the `Basic` tab when you have an API key from Anthropic, OpenAI, Google, Ope
 
 For model recommendations and provider references, see [LLM Configuration](/openhands/usage/llms/llms). For the OpenHands provider, see [OpenHands LLM Provider](/openhands/usage/llms/openhands-llms).
 
+## OCI Generative AI
+
+Use the first-class `OCI Generative AI` provider when you have an OCI Generative AI API key and project. OpenHands uses OCI's OpenAI-compatible Responses API for this provider.
+
+### Before You Begin
+
+1. Choose an [OCI region that supports Generative AI](https://docs.oracle.com/en-us/iaas/Content/generative-ai/regions.htm).
+2. Create or select a [Generative AI project](https://docs.oracle.com/en-us/iaas/Content/generative-ai/projects.htm), then copy its project OCID.
+3. Make sure your OCI identity has the required [project permissions](https://docs.oracle.com/en-us/iaas/Content/generative-ai/project-permissions.htm).
+4. Create a [Generative AI API key](https://docs.oracle.com/en-us/iaas/Content/generative-ai/api-keys.htm).
+5. Choose an exact model ID from the [pretrained models catalog](https://docs.oracle.com/en-us/iaas/Content/generative-ai/pretrained-models.htm). Confirm that the model is available in your selected region by checking [models by region](https://docs.oracle.com/en-us/iaas/Content/generative-ai/model-endpoint-regions.htm).
+
+<Note>
+The initial integration supports OCI Generative AI API keys. OCI IAM or SigV4 authentication is not supported by this provider path.
+</Note>
+
+### Create the Profile
+
+1. Open `Settings > LLM` and select the `Basic` tab.
+2. Select `OCI Generative AI` as the provider.
+3. Enter the exact model ID from OCI. Do not add an `oci_genai/` prefix in the model field; Agent Canvas adds the provider prefix when it saves the profile.
+4. Enter the OCI region identifier, such as `us-chicago-1`.
+5. Enter the Generative AI project OCID. It starts with `ocid1.generativeaiproject.`.
+6. Enter the Generative AI API key.
+7. Save the profile, start a new conversation, and send a short message that requires a tool call.
+
+OpenHands derives this endpoint from the region:
+
+```text
+https://inference.generativeai.<region>.oci.oraclecloud.com/openai/v1
+```
+
+It also selects the Responses API and sends the project OCID with each provider request. You do not need to enter a base URL, custom headers, or capability overrides.
+
+The API key uses the same secret handling as other LLM credentials and is not returned in plaintext by profile APIs. Configure `OH_SECRET_KEY` on the Agent Server to encrypt profile credentials at rest. The region, project OCID, and model ID are non-secret profile settings.
+
+### Credential-Gated Live Test Plan
+
+The automated unit tests cover endpoint derivation, project-header injection, normal Responses API output, and tool-call parsing without making a billable provider request. Before release, maintainers can perform a live check with a temporary OCI API key:
+
+1. Create a profile by following the steps above.
+2. Start a new conversation and ask: `Use the terminal to print the current working directory.`
+3. Confirm that the assistant calls the terminal tool and continues after the tool result.
+4. Remove or rotate the temporary API key after the test.
+
 ## Local OpenAI-Compatible Server
 
 Use the `Advanced` tab for LM Studio, Ollama, vLLM, SGLang, or another server that exposes an OpenAI-compatible API.