feat(examples): Add Local RAG demo

- Add local_rag_demo example demonstrating self-managed RAG pipeline
- Shows SentenceChunker, FixedSizeChunker, and LLM integration
- Documents stay local, only chunks sent for embeddings
- Add aiofiles dependency to agents-core for async file operations

Author: d3xvn

agents-core/pyproject.toml

@@ -27,6 +27,7 @@ dependencies = [
     "numpy>=1.24.0",
     "mcp>=1.16.0",
     "colorlog>=6.10.1",
+    "aiofiles>=24.1.0",
 ]
 
 [project.urls]

examples/other_examples/local_rag_demo/README.md

@@ -0,0 +1,61 @@
+# Local RAG Demo
+
+Demonstrates self-managed RAG with pluggable components.
+
+## What makes it "local"?
+
+Unlike managed RAG (Gemini/OpenAI Vector Store), LocalRAG gives you control:
+
+| Component | What happens |
+|-----------|--------------|
+| **Chunking** | Done locally by your chosen chunker |
+| **Embeddings** | API call returns vectors to you (text not stored remotely) |
+| **Vector Store** | Stored in local memory |
+| **Search** | Local cosine similarity (no API call) |
+
+Your documents never leave your machine - only text chunks are sent to the embedding API.
+
+## Components
+
+- **Chunkers**: `SentenceChunker`, `FixedSizeChunker`
+- **Embeddings**: `OpenAIEmbeddings` (or implement your own)
+- **Vector Stores**: `InMemoryVectorStore` (or implement your own)
+
+## Setup
+
+```bash
+export OPENAI_API_KEY=your-key-here
+```
+
+## Run
+
+```bash
+cd examples/other_examples/local_rag_demo
+uv run python local_rag_demo.py
+```
+
+## Demos included
+
+1. **SentenceChunker** - Natural text boundaries
+2. **FixedSizeChunker** - Fixed size with overlap
+3. **LLM Integration** - Automatic context injection
+4. **File ingestion** - Add files directly
+
+## Making it fully offline
+
+To eliminate all API calls, implement a local embedding provider:
+
+```python
+class LocalEmbeddings(EmbeddingProvider):
+    def __init__(self):
+        from sentence_transformers import SentenceTransformer
+        self._model = SentenceTransformer("all-MiniLM-L6-v2")
+    
+    @property
+    def dimension(self) -> int:
+        return 384
+    
+    async def embed(self, text: str) -> list[float]:
+        return self._model.encode(text).tolist()
+```
+

examples/other_examples/local_rag_demo/__init__.py

@@ -0,0 +1,2 @@
+# Local RAG Demo
+