ollama support updates

.devcontainer/ollama/devcontainer.json

@@ -7,7 +7,9 @@
         "context": "../.."
     },
     "features": {
-        "ghcr.io/prulloac/devcontainer-features/ollama:1": {}
+        "ghcr.io/prulloac/devcontainer-features/ollama:1": {
+            "pull": "gemma4:e4b"
+        }
     },
     // Configure tool-specific properties.
     "customizations": {
@@ -26,7 +28,7 @@
     },
 
     // Use 'postCreateCommand' to run commands after the container is created.
-    "postCreateCommand": "cp .env.sample.ollama .env && ollama pull llama3.1",
+    "postCreateCommand": "cp .env.sample.ollama .env",
 
     // Comment out to connect as root instead. More info: https://aka.ms/vscode-remote/containers/non-root.
     "remoteUser": "vscode",

.env.sample

@@ -5,7 +5,7 @@ AZURE_OPENAI_ENDPOINT=https://YOUR-AZURE-OPENAI-SERVICE-NAME.openai.azure.com
 AZURE_OPENAI_CHAT_DEPLOYMENT=YOUR-AZURE-DEPLOYMENT-NAME
 # Needed for Ollama:
 OLLAMA_ENDPOINT=http://localhost:11434/v1
-OLLAMA_MODEL=llama3.1
+OLLAMA_MODEL=gemma4:e4b
 # Needed for OpenAI.com:
 OPENAI_KEY=YOUR-OPENAI-KEY
 OPENAI_MODEL=gpt-3.5-turbo

.env.sample.ollama

@@ -1,4 +1,4 @@
 # See .env.sample for all options
 API_HOST=ollama
 OLLAMA_ENDPOINT=http://localhost:11434/v1
-OLLAMA_MODEL=llama3.1
+OLLAMA_MODEL=gemma4:e4b

README.md

@@ -142,27 +142,43 @@ This project includes infrastructure as code (IaC) to provision Azure OpenAI dep
 
 ## Using Ollama models
 
+Most chat, streaming, function calling, structured outputs, CSV RAG, and document RAG flow samples work with local Ollama chat models. These samples have been tested with `gemma4:e4b` and `qwen3.5:4b`. The document ingestion and hybrid vector search samples currently use `text-embedding-3-small` for embeddings, so those scripts need Azure OpenAI/OpenAI embeddings or a code update before they can run in a local-only Ollama setup. The `reasoning.py` sample is intended for reasoning models, such as `gpt-oss`.
+
+If you use GitHub Codespaces or Dev Containers, you can use the Ollama devcontainer, which installs Ollama and pulls the default model for you:
+
+```text
+https://codespaces.new/Azure-Samples/python-openai-demos?devcontainer_path=.devcontainer/ollama/devcontainer.json
+```
+
 1. Install [Ollama](https://ollama.com/) and follow the instructions to set it up on your local machine.
-2. Pull a model, for example:
+2. Pull the recommended model:
+
+    ```shell
+    ollama pull gemma4:e4b
+    ```
+
+    Another tested option is:
 
     ```shell
-    ollama pull llama3.1
+    ollama pull qwen3.5:4b
     ```
 
-3. Create a `.env` file by copying the `.env.sample` file and updating it with your Ollama endpoint and model name.
+3. Create a `.env` file by copying the Ollama-specific environment sample:
 
     ```bash
-    cp .env.sample .env
+    cp .env.sample.ollama .env
     ```
 
-4. Update the `.env` file with your Ollama endpoint and model name (any model you've pulled):
+4. Update the `.env` file with your Ollama endpoint and model name, if needed:
 
     ```bash
     API_HOST=ollama
     OLLAMA_ENDPOINT=http://localhost:11434/v1
-    OLLAMA_MODEL=llama3.1
+    OLLAMA_MODEL=gemma4:e4b
     ```
 
+Use `http://localhost:11434/v1` when Ollama and Python run in the same environment, including the Ollama devcontainer. If Python runs in a different container and Ollama runs on the host machine, use `http://host.docker.internal:11434/v1` instead.
+
 ## Resources
 
 * [Video series: Learn Python + AI (October 2025)](https://techcommunity.microsoft.com/blog/educatordeveloperblog/level-up-your-python--ai-skills-with-our-complete-series/4464546)

requirements-rag.txt

@@ -1,3 +1,4 @@
 pymupdf4llm
 lunr
 sentence-transformers
+tiktoken

spanish/README.md

@@ -137,22 +137,30 @@ OPENAI_MODEL=gpt-4o-mini
 
 ### Usando modelos de Ollama
 
-Instala [Ollama](https://ollama.com/) y descarga un modelo:
+Instala [Ollama](https://ollama.com/) y descarga el modelo recomendado:
 
 ```bash
-ollama pull llama3.1
+ollama pull gemma4:e4b
+```
+
+Otra opcion probada es:
+
+```bash
+ollama pull qwen3.5:4b
 ```
 
 Configura tu `.env`:
 
 ```bash
 API_HOST=ollama
 OLLAMA_ENDPOINT=http://localhost:11434/v1
-OLLAMA_MODEL=llama3.1
+OLLAMA_MODEL=gemma4:e4b
 ```
 
 Si ejecutas dentro de un Dev Container, reemplaza `localhost` por `host.docker.internal`.
 
+La mayoria de los ejemplos de chat, streaming, function calling, salidas estructuradas, RAG con CSV y flujo RAG con documentos funcionan con modelos de chat locales de Ollama. Los ejemplos de ingesta de documentos y busqueda vectorial hibrida actualmente usan `text-embedding-3-small` para embeddings, asi que esos scripts necesitan embeddings de Azure OpenAI/OpenAI o una actualizacion de codigo antes de poder ejecutarse en una configuracion local solo con Ollama.
+
 ## Recursos
 
 * [Próxima serie octubre 2025: Python + IA](https://aka.ms/PythonIA/serie)