Add OCI LLM, embeddings, and Oracle integrations

docs/en/concepts/llms.mdx

@@ -772,6 +772,224 @@ In this section, you'll find detailed examples that help you select, configure,
     ```
   </Accordion>
 
+  <Accordion title="Oracle Cloud Infrastructure (OCI)">
+    CrewAI provides native integration with OCI Generative AI for generic chat models, OpenAI-hosted OCI models, Google-hosted OCI models, Meta-hosted OCI models, and dedicated inference endpoints.
+
+    **Recommended Cohere Models (verified in OCI on March 10, 2026):**
+    - `cohere.command-a-reasoning` for reasoning-heavy text workflows
+    - `cohere.command-a-03-2025` for general text generation and streaming
+    - `cohere.command-a-vision` for the top Cohere multimodal tier once OCI Cohere vision formatting is enabled in CrewAI
+
+    **Recommended Regions for Cohere in OCI:**
+    - choose any subscribed OCI region where the target model is available
+    - common examples include `us-chicago-1`, `us-ashburn-1`, `uk-london-1`, and `eu-paris-1`
+
+    ```toml Code
+    # Required
+    OCI_COMPARTMENT_ID=ocid1.compartment.oc1..exampleuniqueID
+
+    # Optional when not passing service_endpoint directly
+    OCI_REGION=<your-oci-region>
+
+    # Authentication options
+    OCI_AUTH_TYPE=API_KEY
+    OCI_AUTH_PROFILE=DEFAULT
+    OCI_AUTH_FILE_LOCATION=~/.oci/config
+
+    # Optional explicit endpoint override
+    OCI_SERVICE_ENDPOINT=https://inference.generativeai.<your-oci-region>.oci.oraclecloud.com
+    ```
+
+    **Basic Usage:**
+    ```python Code
+    from crewai import LLM
+
+    llm = LLM(
+        model="oci/cohere.command-a-reasoning",
+        compartment_id="ocid1.compartment.oc1..exampleuniqueID",
+        auth_type="API_KEY",
+        auth_profile="DEFAULT",
+        temperature=0,
+        max_tokens=512,
+    )
+    ```
+
+    **Provider Routing Examples:**
+    ```python Code
+    from crewai import LLM
+
+    meta_llm = LLM(
+        model="oci/meta.llama-3.3-70b-instruct",
+        compartment_id="ocid1.compartment.oc1..exampleuniqueID",
+    )
+
+    gemini_llm = LLM(
+        model="oci/google.gemini-2.5-flash",
+        compartment_id="ocid1.compartment.oc1..exampleuniqueID",
+    )
+
+    openai_llm = LLM(
+        model="oci/openai.gpt-4o-mini",
+        compartment_id="ocid1.compartment.oc1..exampleuniqueID",
+    )
+    ```
+
+    **Async Usage:**
+    ```python Code
+    import asyncio
+    from crewai import LLM
+
+    llm = LLM(
+        model="oci/cohere.command-a-03-2025",
+        compartment_id="ocid1.compartment.oc1..exampleuniqueID",
+    )
+
+    async def main():
+        result = await llm.acall("Summarize Oracle Cloud in one sentence.")
+        print(result)
+
+    asyncio.run(main())
+    ```
+
+    **Streaming Usage:**
+    ```python Code
+    from crewai import LLM
+
+    llm = LLM(
+        model="oci/cohere.command-a-03-2025",
+        compartment_id="ocid1.compartment.oc1..exampleuniqueID",
+        stream=True,
+        temperature=0,
+    )
+
+    result = llm.call("Reply with exactly three words about Oracle Cloud.")
+    print(result)
+    ```
+
+    **Multimodal Usage:**
+    ```python Code
+    from crewai import LLM
+    from crewai_files import ImageFile
+
+    llm = LLM(
+        model="oci/google.gemini-2.5-flash",
+        compartment_id="ocid1.compartment.oc1..exampleuniqueID",
+    )
+
+    response = llm.call(
+        [
+            {
+                "role": "user",
+                "content": "Describe this architecture diagram.",
+                "files": {
+                    "diagram": ImageFile(source="./architecture.png"),
+                },
+            }
+        ]
+    )
+
+    print(response)
+    ```
+
+    **Structured Outputs:**
+    ```python Code
+    from pydantic import BaseModel
+    from crewai import LLM
+
+    class OCIAnswer(BaseModel):
+        topic: str
+        summary: str
+
+    llm = LLM(
+        model="oci/cohere.command-a-reasoning",
+        compartment_id="ocid1.compartment.oc1..exampleuniqueID",
+        temperature=0,
+    )
+
+    response = llm.call(
+        "Return a short JSON summary about Oracle Cloud.",
+        response_model=OCIAnswer,
+    )
+    print(response.summary)
+    ```
+
+    **Agent + Tool Usage:**
+    ```python Code
+    from crewai import Agent, LLM
+    from crewai.tools import tool
+
+    @tool
+    def add_numbers(a: int, b: int) -> int:
+        """Add two numbers and return the sum."""
+        return a + b
+
+    agent = Agent(
+        role="OCI Calculator",
+        goal="Use tools to solve arithmetic problems",
+        backstory="You are a precise calculator assistant.",
+        llm=LLM(
+            model="oci/cohere.command-a-03-2025",
+            compartment_id="ocid1.compartment.oc1..exampleuniqueID",
+            temperature=0,
+        ),
+        tools=[add_numbers],
+        verbose=True,
+    )
+
+    result = agent.kickoff(
+        "Use the add_numbers tool to calculate 15 + 27. Return only the final result."
+    )
+    print(result.raw)
+    ```
+
+    **Dedicated Endpoint Usage:**
+    ```python Code
+    from crewai import LLM
+
+    llm = LLM(
+        model="oci/ocid1.generativeaiendpoint.oc1.<your-oci-region>.exampleuniqueID",
+        compartment_id="ocid1.compartment.oc1..exampleuniqueID",
+    )
+    ```
+
+    **Supported Authentication Types:**
+    - `API_KEY`
+    - `SECURITY_TOKEN`
+    - `INSTANCE_PRINCIPAL`
+    - `RESOURCE_PRINCIPAL`
+
+    **Features:**
+    - Native function calling support
+    - Structured outputs with JSON schema response formats
+    - Async `acall()` support
+    - Streaming responses for OCI chat models
+    - Multimodal generic chat inputs for text, images, documents, video, and audio
+    - Stop sequences support
+    - Dedicated endpoint routing
+    - Generic chat support for OCI-hosted OpenAI, Google, and Meta models
+    - Cohere-specific OCI chat formatting for text models
+    - OCI embeddings in CrewAI RAG via the `oci` embedding provider
+    - CrewAI-managed OCI retrieval via `/en/tools/cloud-storage/ociknowledgebasetool`
+
+    **Current Limitations:**
+    - Cohere multimodal and vision-specific formatting is not implemented yet
+
+    **`langchain-oci` Sample Coverage in CrewAI:**
+    - `01-getting-started`: native OCI auth, provider routing, and streaming examples above
+    - `02-vision-and-multimodal`: supported through CrewAI file attachments on OCI multimodal models
+    - `03-building-ai-agents`: maps directly to CrewAI `Agent` usage with an OCI-backed `LLM`
+    - `04-tool-calling-mastery`: supported through CrewAI tools and OCI native function calling
+    - `05-structured-output`: supported through `response_model`
+    - `07-async-for-production`: supported through `acall()`
+    - `09-provider-deep-dive`: covered by provider-prefixed model routing and dedicated endpoint support
+    - `10-embeddings`: covered by the OCI embedding provider for text and image embeddings, plus [`OCIKnowledgeBaseTool`](/en/tools/cloud-storage/ociknowledgebasetool) for CrewAI-managed retrieval
+
+    **Install:**
+    ```bash
+    uv add "crewai[oci]"
+    ```
+  </Accordion>
+
   <Accordion title="Amazon SageMaker">
     ```toml Code
     AWS_ACCESS_KEY_ID=<your-access-key>

docs/en/tools/cloud-storage/ociknowledgebasetool.mdx

@@ -0,0 +1,111 @@
+---
+title: "OCI Knowledge Base Tool"
+description: "Build and query a CrewAI-managed knowledge base powered by OCI embeddings"
+icon: "database"
+mode: "wide"
+---
+
+# `OCIKnowledgeBaseTool`
+
+The `OCIKnowledgeBaseTool` gives your agents a CrewAI-managed retrieval tool powered by OCI embedding models. It uses CrewAI's native `RagTool` stack and defaults to OCI embeddings, so you can load documents, directories, or URLs and query them semantically from your crews.
+
+This is the closest CrewAI-native OCI equivalent to the Bedrock knowledge-base workflow. Unlike Amazon Bedrock Knowledge Bases, the index is managed inside CrewAI's RAG system rather than an OCI managed retrieval service.
+
+## Installation
+
+```bash
+uv add "crewai[oci]"
+uv add "crewai-tools[oci]"
+```
+
+## Example
+
+```python
+from crewai import Agent
+from crewai_tools import OCIKnowledgeBaseTool
+
+kb_tool = OCIKnowledgeBaseTool(
+    knowledge_source="./oracle-architecture.pdf",
+    compartment_id="ocid1.compartment.oc1..exampleuniqueID",
+    region="<your-oci-region>",
+)
+
+agent = Agent(
+    role="OCI Research Analyst",
+    goal="Answer architecture questions with the indexed OCI knowledge base",
+    tools=[kb_tool],
+    verbose=True,
+)
+```
+
+## Add Sources Dynamically
+
+```python
+from crewai_tools import OCIKnowledgeBaseTool
+
+kb_tool = OCIKnowledgeBaseTool(
+    compartment_id="ocid1.compartment.oc1..exampleuniqueID",
+    region="<your-oci-region>",
+)
+
+kb_tool.add("./runbooks/networking.md")
+kb_tool.add("./runbooks/security/")
+kb_tool.add("https://docs.oracle.com/en-us/iaas/Content/home.htm")
+```
+
+## Configuration
+
+The tool defaults to this embedding configuration:
+
+```python
+{
+    "embedding_model": {
+        "provider": "oci",
+        "config": {
+            "model_name": "cohere.embed-english-v3.0",
+            "compartment_id": "ocid1.compartment.oc1..exampleuniqueID",
+            "region": "<your-oci-region>",
+            "auth_type": "API_KEY",
+            "auth_profile": "DEFAULT",
+            "auth_file_location": "~/.oci/config",
+        }
+    }
+}
+```
+
+You can override the vector database layer with standard `RagTool` config:
+
+```python
+from crewai_tools import OCIKnowledgeBaseTool
+
+kb_tool = OCIKnowledgeBaseTool(
+    compartment_id="ocid1.compartment.oc1..exampleuniqueID",
+    config={
+        "vectordb": {
+            "provider": "qdrant",
+            "config": {
+                "url": "http://localhost:6333",
+                "api_key": "qdrant-key",
+            },
+        }
+    },
+)
+```
+
+## Environment Variables
+
+```bash
+OCI_COMPARTMENT_ID=ocid1.compartment.oc1..exampleuniqueID
+OCI_REGION=<your-oci-region>
+OCI_AUTH_TYPE=API_KEY
+OCI_AUTH_PROFILE=DEFAULT
+OCI_AUTH_FILE_LOCATION=~/.oci/config
+OCI_EMBED_MODEL=cohere.embed-english-v3.0
+```
+
+## Notes
+
+- Uses CrewAI's native RAG stack, not an OCI managed knowledge-base service
+- Supports any source type that `RagTool` can ingest
+- Defaults to OCI embeddings, but you can still override the `config` field for advanced vector store configuration
+- For direct OCI text or image embedding workflows with `cohere.embed-v4.0`, use the CrewAI OCI embedding provider outside of `OCIKnowledgeBaseTool`

docs/en/tools/cloud-storage/ociobjectstoragereadertool.mdx

@@ -0,0 +1,38 @@
+---
+title: "OCI Object Storage Reader Tool"
+description: "Read files from Oracle Cloud Infrastructure Object Storage"
+icon: "cloud"
+mode: "wide"
+---
+
+# `OCIObjectStorageReaderTool`
+
+Use `OCIObjectStorageReaderTool` to read text files from Oracle Cloud Infrastructure Object Storage inside a CrewAI workflow.
+
+## Installation
+
+```bash
+uv pip install 'crewai-tools[oci]'
+```
+
+## Usage
+
+```python
+from crewai import Agent
+from crewai_tools import OCIObjectStorageReaderTool
+
+oci_reader = OCIObjectStorageReaderTool(namespace_name="my-namespace")
+
+agent = Agent(
+    role="Cloud Reader",
+    goal="Fetch cloud-hosted files",
+    tools=[oci_reader],
+)
+```
+
+## Path Formats
+
+- `oci://bucket/path/to/file.txt`
+- `oci://namespace@bucket/path/to/file.txt`
+
+If the namespace is omitted, the tool will use `namespace_name`, `OCI_OBJECT_STORAGE_NAMESPACE`, or fetch the namespace from OCI automatically.