MacOS LLM Controller lets users execute macOS commands using natural language via text or voice. It displays a real-time status and command history. Uses LlamaStack, React, and Flask. See setup steps in readme.

Sally O'Malley explains the unique observability challenges of LLMs and provides a reproducible, open-source stack for monitoring AI workloads. She demonstrates deploying Prometheus, Grafana, OpenTele...

Explore how to run tools with Node.js using Llama Stack's completions API, agent API, and support for in-line tools, local MCP tools, and remote MCP tools.

This tutorial shows you how to use the Llama Stack API to implement retrieval-augmented generation for an AI application built with Node.js

Retrieval Augmented Generation (RAG) has quickly become a popular choice for building intelligent AI applications, yet taking one from concept to production can often be complex. This workshop aims to...

Explore how to run tools with Node.js using Llama Stack's completions API, agent API, and support for in-line tools, local MCP tools, and remote MCP tools.

github.com/llamastack/llama-stack/pull/3969/files They added access control logic in get_session_info and get_session_if_accessible: if user is None They skip certain checks. That means behaviors for anonymous user might diverge. While the type safety improves, sometimes stricter types can force runtime changes or reveal latent bugs

Shoutout Llamastack x 8VC challenge - WashU This is a React and Flask-based app that processes natural language commands to execute macOS system actions using Python code generated by an LLM.

Shoutout Llamastack x 8VC challenge - WashU This is a React and Flask-based app that processes natural language commands to execute macOS system actions using Python code generated by an LLM.

At MWC 2026 in Barcelona, as Europe wrestles with AI sovereignty and hyperscaler dominance, Telenor and Red Hat unveiled a collaboration for a Norway-based AI factory aimed at offering companies high-performance AI compute—without ceding control to foreign cloud providers. Telenor is using Red Hat OpenShift AI as its main environment to build, train, and run AI agents and applications, including RAG and agentic workflows with LlamaStack. Together with Red Hat OpenShift Platform Plus, it serves as the cloud‑native base layer for the AI factory. The AI factory is designed to provide high-performance compute for AI models while keeping data inside Norway, so companies can meet EU data sovereignty rules without having to trade away speed or capacity. “With the pace of change in the AI market, Red Hat and Telenor AI Factory are catering for today’s greatest strategic need: The freedom to choose any model, on any accelerator, across any environment,” said Rich Stephens, Red Hat’s vice president, EMEA Telecommunications. “Together we are delivering support for sovereignty, governance and enhanced systems security so that customers can scale AI efforts to drive value, with control and autonomy over data.” A Sovereign Advantage? Telenor’s AI Factory runs on data center infrastructure located...

Red Hat Brings Distributed AI Inference to Production AI Workloads with Red Hat AI 3

<p>We are excited to announce that vLLM inference provider is now available in <a href="https://github.com/meta-llama/llama-stack">Llama Stack</a> through the collaboration between the Red Hat AI Engineering team and the Llama Stack team from Meta. This article provides an introduction to this integration and a tutorial to help you get started using it locally or deploying it in a Kubernetes cluster.</p> <h1 id="what-is-llama-stack">What is Llama Stack?</h1> <p><img align="right" alt="llama-stack-diagram" height="50%" src="https://blog.vllm.ai/assets/figures/llama-stack/llama-stack.png" width="50%"/></p> <p>Llama Stack defines and standardizes the set of core building blocks needed to bring generative AI applications to market. These building blocks are presented in the form of interoperable APIs with a broad set of Service Providers providing their implementations.</p> <p>Llama Stack focuses on making it easy to build production applications with a variety of models - ranging from the latest Llama 3.3 model to specialized models like Llama Guard for safety and other models. The goal is to provide pre-packaged implementations (aka “distributions”) which can be run in a variety of deployment environments. The Stack can assist you in your entire app development lifecycle - start iterating on local, mobile or desktop and seamlessly transition to on-prem or public cloud deployments. At every point in this transition, the same set of APIs and the same developer experience are available.</p> <p>Each specific implementation of an API is called a “Provider” in this architecture. Users can swap providers via configuration. vLLM is a prominent example of a high-performance API backing the inference API.</p> <h1 id="vllm-inference-provider">vLLM Inference Provider</h1> <p>Llama Stack provides two vLLM inference providers:</p> <ol> <li><a href="https://llama-stack.readthedocs.io/en/latest/distributions/self_hosted_distro/remote-vllm.html">Remote vLLM inference provider</a> through vLLM’s <a href="https://docs.vllm.ai/en/latest/getting_started/quickstart.html#openai-completions-api-with-vllm">OpenAI-compatible server</a>;</li> <li><a href="https://github.com/meta-llama/llama-stack/tree/main/llama_stack/providers/inline/inference/vllm">Inline vLLM inference provider</a> that runs alongside with Llama Stack server.</li> </ol> <p>In this article, we will demonstrate the functionality through the remote vLLM inference provider.</p> <h1 id="tutorial">Tutorial</h1> <h2 id="prerequisites">Prerequisites</h2> <ul> <li>Linux operating system</li> <li><a href="https://huggingface.co/docs/huggingface_hub/main/en/guides/cli">Hugging Face CLI</a> if you’d like to download the model via CLI.</li> <li>OCI-compliant container technologies like <a href="https://podman.io/">Podman</a> or <a href="https://www.docker.com/">Docker</a> (can be specified via the <code class="language-plaintext highlighter-rouge">CONTAINER_BINARY</code> environment variable when running <code class="language-plaintext highlighter-rouge">llama stack</code> CLI commands).</li> <li><a href="https://kind.sigs.k8s.io/">Kind</a> for Kubernetes deployment.</li> <li><a href="https://github.com/conda/conda">Conda</a> for managing Python environment.</li> </ul> <h2 id="get-started-via-containers">Get Started via Containers</h2> <h3 id="start-vllm-server">Start vLLM Server</h3> <p>We first download the “Llama-3.2-1B-Instruct” model using the <a href="https://huggingface.co/docs/huggingface_hub/main/en/guides/cli">Hugging Face CLI</a>. Note that you’ll need to specify your Hugging Face token when logging in.</p> <div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="nb">mkdir</span> /tmp/test-vllm-llama-stack huggingface-cli login <span class="nt">--token</span> &lt;YOUR-HF-TOKEN&gt; huggingface-cli download meta-llama/Llama-3.2-1B-Instruct <span class="nt">--local-dir</span> /tmp/test-vllm-llama-stack/.cache/huggingface/hub/models/Llama-3.2-1B-Instruct </code></pre></div></div> <p>Next, let’s build the vLLM CPU container image from source. Note that while we use it for demonstration purposes, there are plenty of <a href="https://docs.vllm.ai/en/latest/getting_started/installation/index.html">other images available for different hardware and architectures</a>.</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>git clone git@github.com:vllm-project/vllm.git /tmp/test-vllm-llama-stack cd /tmp/test-vllm-llama-stack/vllm podman build -f Dockerfile.cpu -t vllm-cpu-env --shm-size=4g . </code></pre></div></div> <p>We can then start the vLLM container:</p> <div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code>podman run <span class="nt">-it</span> <span class="nt">--network</span><span class="o">=</span>host <span class="se">\</span> <span class="nt">--group-add</span><span class="o">=</span>video <span class="se">\</span> <span class="nt">--ipc</span><span class="o">=</span>host <span class="se">\</span> <span class="nt">--cap-add</span><span class="o">=</span>SYS_PTRACE <span class="se">\</span> <span class="nt">--security-opt</span> <span class="nv">seccomp</span><span class="o">=</span>unconfined <span class="se">\</span> <span class="nt">--device</span> /dev/kfd <span class="se">\</span> <span class="nt">--device</span> /dev/dri <span class="se">\</span> <span class="nt">-v</span> /tmp/test-vllm-llama-stack/.cache/huggingface/hub/models/Llama-3.2-1B-Instruct:/app/model <span class="se">\</span> <span class="nt">--entrypoint</span><span class="o">=</span><span class="s1">'["python3", "-m", "vllm.entrypoints.openai.api_server", "--model", "/app/model", "--served-model-name", "meta-llama/Llama-3.2-1B-Instruct", "--port", "8000"]'</span> <span class="se">\</span> vllm-cpu-env </code></pre></div></div> <p>We can get a list of models and test a prompt once the model server has started:</p> <div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code>curl http://localhost:8000/v1/models curl http://localhost:8000/v1/completions <span class="se">\</span> <span class="nt">-H</span> <span class="s2">"Content-Type: application/json"</span> <span class="se">\</span> <span class="nt">-d</span> <span class="s1">'{ "model": "meta-llama/Llama-3.2-1B-Instruct", "prompt": "San Francisco is a", "max_tokens": 7, "temperature": 0 }'</span> </code></pre></div></div> <h3 id="start-llama-stack-server">Start Llama Stack Server</h3> <p>Once we verify that the vLLM server has started successfully and is able to serve requests, we can then build and start the Llama Stack server.</p> <p>First, we clone the Llama Stack source code and create a Conda environment that includes all the dependencies:</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>git clone git@github.com:meta-llama/llama-stack.git /tmp/test-vllm-llama-stack/llama-stack cd /tmp/test-vllm-llama-stack/llama-stack conda create -n stack python=3.10 conda activate stack pip install . </code></pre></div></div> <p>Next, we build the container image with <code class="language-plaintext highlighter-rouge">llama stack build</code>:</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>cat &gt; /tmp/test-vllm-llama-stack/vllm-llama-stack-build.yaml &lt;&lt; "EOF" name: vllm distribution_spec: description: Like local, but use vLLM for running LLM inference providers: inference: remote::vllm safety: inline::llama-guard agents: inline::meta-reference vector_io: inline::faiss datasetio: inline::localfs scoring: inline::basic eval: inline::meta-reference post_training: inline::torchtune telemetry: inline::meta-reference image_type: container EOF export CONTAINER_BINARY=podman LLAMA_STACK_DIR=. PYTHONPATH=. python -m llama_stack.cli.llama stack build --config /tmp/test-vllm-llama-stack/vllm-llama-stack-build.yaml --image-name distribution-myenv </code></pre></div></div> <p>Once the container image has been built successfully, we can then edit the generated <code class="language-plaintext highlighter-rouge">vllm-run.yaml</code> to be <code class="language-plaintext highlighter-rouge">/tmp/test-vllm-llama-stack/vllm-llama-stack-run.yaml</code> with the following change in the <code class="language-plaintext highlighter-rouge">models</code> field:</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>models: - metadata: {} model_id: ${env.INFERENCE_MODEL} provider_id: vllm provider_model_id: null </code></pre></div></div> <p>Then we can start the LlamaStack Server with the image we built via <code class="language-plaintext highlighter-rouge">llama stack run</code>:</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>export INFERENCE_ADDR=host.containers.internal export INFERENCE_PORT=8000 export INFERENCE_MODEL=meta-llama/Llama-3.2-1B-Instruct export LLAMA_STACK_PORT=5000 LLAMA_STACK_DIR=. PYTHONPATH=. python -m llama_stack.cli.llama stack run \ --env INFERENCE_MODEL=$INFERENCE_MODEL \ --env VLLM_URL=http://$INFERENCE_ADDR:$INFERENCE_PORT/v1 \ --env VLLM_MAX_TOKENS=8192 \ --env VLLM_API_TOKEN=fake \ --env LLAMA_STACK_PORT=$LLAMA_STACK_PORT \ /tmp/test-vllm-llama-stack/vllm-llama-stack-run.yaml </code></pre></div></div> <p>Alternatively, we can run the following <code class="language-plaintext highlighter-rouge">podman run</code> command instead:</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>podman run --security-opt label=disable -it --network host -v /tmp/test-vllm-llama-stack/vllm-llama-stack-run.yaml:/app/config.yaml -v /tmp/test-vllm-llama-stack/llama-stack:/app/llama-stack-source \ --env INFERENCE_MODEL=$INFERENCE_MODEL \ --env VLLM_URL=http://$INFERENCE_ADDR:$INFERENCE_PORT/v1 \ --env VLLM_MAX_TOKENS=8192 \ --env VLLM_API_TOKEN=fake \ --env LLAMA_STACK_PORT=$LLAMA_STACK_PORT \ --entrypoint='["python", "-m", "llama_stack.distribution.server.server", "--yaml-config", "/app/config.yaml"]' \ localhost/distribution-myenv:dev </code></pre></div></div> <p>Once we start the Llama Stack server successfully, we can then start testing a inference request:</p> <p>Via Bash:</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>llama-stack-client --endpoint http://localhost:5000 inference chat-completion --message "hello, what model are you?" </code></pre></div></div> <p>Output:</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>ChatCompletionResponse( completion_message=CompletionMessage( content="Hello! I'm an AI, a conversational AI model. I'm a type of computer program designed to understand and respond to human language. My creators have trained me on a vast amount of text data, allowing me to generate human-like responses to a wide range of questions and topics. I'm here to help answer any question you may have, so feel free to ask me anything!", role='assistant', stop_reason='end_of_turn', tool_calls=[] ), logprobs=None ) </code></pre></div></div> <p>Via Python:</p> <div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="kn">import</span> <span class="n">os</span> <span class="kn">from</span> <span class="n">llama_stack_client</span> <span class="kn">import</span> <span class="n">LlamaStackClient</span> <span class="n">client</span> <span class="o">=</span> <span class="nc">LlamaStackClient</span><span class="p">(</span><span class="n">base_url</span><span class="o">=</span><span class="sa">f</span><span class="sh">"</span><span class="s">http://localhost:</span><span class="si">{</span><span class="n">os</span><span class="p">.</span><span class="n">environ</span><span class="p">[</span><span class="sh">'</span><span class="s">LLAMA_STACK_PORT</span><span class="sh">'</span><span class="p">]</span><span class="si">}</span><span class="sh">"</span><span class="p">)</span> <span class="c1"># List available models </span><span class="n">models</span> <span class="o">=</span> <span class="n">client</span><span class="p">.</span><span class="n">models</span><span class="p">.</span><span class="nf">list</span><span class="p">()</span> <span class="nf">print</span><span class="p">(</span><span class="n">models</span><span class="p">)</span> <span class="n">response</span> <span class="o">=</span> <span class="n">client</span><span class="p">.</span><span class="n">inference</span><span class="p">.</span><span class="nf">chat_completion</span><span class="p">(</span> <span class="n">model_id</span><span class="o">=</span><span class="n">os</span><span class="p">.</span><span class="n">environ</span><span class="p">[</span><span class="sh">"</span><span class="s">INFERENCE_MODEL</span><span class="sh">"</span><span class="p">],</span> <span class="n">messages</span><span class="o">=</span><span class="p">[</span> <span class="p">{</span><span class="sh">"</span><span class="s">role</span><span class="sh">"</span><span class="p">:</span> <span class="sh">"</span><span class="s">system</span><span class="sh">"</span><span class="p">,</span> <span class="sh">"</span><span class="s">content</span><span class="sh">"</span><span class="p">:</span> <span class="sh">"</span><span class="s">You are a helpful assistant.</span><span class="sh">"</span><span class="p">},</span> <span class="p">{</span><span class="sh">"</span><span class="s">role</span><span class="sh">"</span><span class="p">:</span> <span class="sh">"</span><span class="s">user</span><span class="sh">"</span><span class="p">,</span> <span class="sh">"</span><span class="s">content</span><span class="sh">"</span><span class="p">:</span> <span class="sh">"</span><span class="s">Write a haiku about coding</span><span class="sh">"</span><span class="p">}</span> <span class="p">]</span> <span class="p">)</span> <span class="nf">print</span><span class="p">(</span><span class="n">response</span><span class="p">.</span><span class="n">completion_message</span><span class="p">.</span><span class="n">content</span><span class="p">)</span> </code></pre></div></div> <p>Output:</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>[Model(identifier='meta-llama/Llama-3.2-1B-Instruct', metadata={}, api_model_type='llm', provider_id='vllm', provider_resource_id='meta-llama/Llama-3.2-1B-Instruct', type='model', model_type='llm')] Here is a haiku about coding: Columns of code flow Logic codes the endless night Tech's silent dawn rise </code></pre></div></div> <h2 id="deployment-on-kubernetes">Deployment on Kubernetes</h2> <p>Instead of starting the Llama Stack and vLLM servers locally. We can deploy them in a Kubernetes cluster. We’ll use a local Kind cluster for demonstration purposes:</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>kind create cluster --image kindest/node:v1.32.0 --name llama-stack-test </code></pre></div></div> <p>Start vLLM server as a Kubernetes Pod and Service (remember to replace <code class="language-plaintext highlighter-rouge">&lt;YOUR-HF-TOKEN&gt;</code> with your actual token):</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>cat &lt;&lt;EOF |kubectl apply -f - apiVersion: v1 kind: PersistentVolumeClaim metadata: name: vllm-models spec: accessModes: - ReadWriteOnce volumeMode: Filesystem resources: requests: storage: 50Gi --- apiVersion: v1 kind: Secret metadata: name: hf-token-secret type: Opaque data: token: "&lt;YOUR-HF-TOKEN&gt;" --- apiVersion: v1 kind: Pod metadata: name: vllm-server labels: app: vllm spec: containers: - name: llama-stack image: localhost/vllm-cpu-env:latest command: - bash - -c - | MODEL="meta-llama/Llama-3.2-1B-Instruct" MODEL_PATH=/app/model/$(basename $MODEL) huggingface-cli login --token $HUGGING_FACE_HUB_TOKEN huggingface-cli download $MODEL --local-dir $MODEL_PATH --cache-dir $MODEL_PATH python3 -m vllm.entrypoints.openai.api_server --model $MODEL_PATH --served-model-name $MODEL --port 8000 ports: - containerPort: 8000 volumeMounts: - name: llama-storage mountPath: /app/model env: - name: HUGGING_FACE_HUB_TOKEN valueFrom: secretKeyRef: name: hf-token-secret key: token volumes: - name: llama-storage persistentVolumeClaim: claimName: vllm-models --- apiVersion: v1 kind: Service metadata: name: vllm-server spec: selector: app: vllm ports: - port: 8000 targetPort: 8000 type: NodePort EOF </code></pre></div></div> <p>We can verify that the vLLM server has started successfully via the logs (this might take a couple of minutes to download the model):</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>$ kubectl logs vllm-server ... INFO: Started server process [1] INFO: Waiting for application startup. INFO: Application startup complete. INFO: Uvicorn running on http://0.0.0.0:8000 (Press CTRL+C to quit) </code></pre></div></div> <p>Then we can modify the previously created <code class="language-plaintext highlighter-rouge">vllm-llama-stack-run.yaml</code> to <code class="language-plaintext highlighter-rouge">/tmp/test-vllm-llama-stack/vllm-llama-stack-run-k8s.yaml</code> with the following inference provider:</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>providers: inference: - provider_id: vllm provider_type: remote::vllm config: url: ${env.VLLM_URL} max_tokens: ${env.VLLM_MAX_TOKENS:4096} api_token: ${env.VLLM_API_TOKEN:fake} </code></pre></div></div> <p>Once we have defined the run configuration for Llama Stack, we can build an image with that configuration the server source code:</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>cat &gt;/tmp/test-vllm-llama-stack/Containerfile.llama-stack-run-k8s &lt;&lt;EOF FROM distribution-myenv:dev RUN apt-get update &amp;&amp; apt-get install -y git RUN git clone https://github.com/meta-llama/llama-stack.git /app/llama-stack-source ADD ./vllm-llama-stack-run-k8s.yaml /app/config.yaml EOF podman build -f /tmp/test-vllm-llama-stack/Containerfile.llama-stack-run-k8s -t llama-stack-run-k8s /tmp/test-vllm-llama-stack </code></pre></div></div> <p>We can then start the LlamaStack server by deploying a Kubernetes Pod and Service:</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>cat &lt;&lt;EOF |kubectl apply -f - apiVersion: v1 kind: PersistentVolumeClaim metadata: name: llama-pvc spec: accessModes: - ReadWriteOnce resources: requests: storage: 1Gi --- apiVersion: v1 kind: Pod metadata: name: llama-stack-pod labels: app: llama-stack spec: containers: - name: llama-stack image: localhost/llama-stack-run-k8s:latest imagePullPolicy: IfNotPresent command: ["python", "-m", "llama_stack.distribution.server.server", "--yaml-config", "/app/config.yaml"] ports: - containerPort: 5000 volumeMounts: - name: llama-storage mountPath: /root/.llama volumes: - name: llama-storage persistentVolumeClaim: claimName: llama-pvc --- apiVersion: v1 kind: Service metadata: name: llama-stack-service spec: selector: app: llama-stack ports: - protocol: TCP port: 5000 targetPort: 5000 type: ClusterIP EOF </code></pre></div></div> <p>We can check that the LlamaStack server has started:</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>$ kubectl logs vllm-server ... INFO: Started server process [1] INFO: Waiting for application startup. INFO: ASGI 'lifespan' protocol appears unsupported. INFO: Application startup complete. INFO: Uvicorn running on http://['::', '0.0.0.0']:5000 (Press CTRL+C to quit) </code></pre></div></div> <p>Now let’s forward the Kubernetes service to a local port and test some inference requests against it via the Llama Stack Client:</p> <div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>kubectl port-forward service/llama-stack-service 5000:5000 llama-stack-client --endpoint http://localhost:5000 inference chat-completion --message "hello, what model are you?" </code></pre></div></div> <p>You can learn more about different providers and functionalities of Llama Stack on <a href="https://llama-stack.readthedocs.io">the official documentation</a>.</p> <h2 id="acknowledgement">Acknowledgement</h2> <p>We’d like to thank the Red Hat AI Engineering team for the implementation of the vLLM inference providers, contributions to many bug fixes, improvements, and key design discussions. We also want to thank the Llama Stack team from Meta and the vLLM team for their timely PR reviews and bug fixes.</p>

YouTube video by Red Hat

Red Hat Brings Distributed AI Inference to Production AI Workloads with Red Hat AI 3

We are excited to announce that vLLM inference provider is now available in Llama Stack through the collaboration between the Red Hat AI Engineering team and the Llama Stack team from Meta. This artic...

Model components of the Llama Stack APIs. Contribute to meta-llama/llama-stack development by creating an account on GitHub.