Integration guide
Node-RED Integration
Flow-based programming tool for wiring devices, APIs, services, and local automation systems.
Recommended path
Node-RED local AI quick start
- 1. Start Ollama and verify the model endpointRun Ollama, pull the chat or embedding model the flow needs, and confirm that http://localhost:11434 responds before opening the Node-RED editor.
- 2. Choose a community node or a direct HTTP callInstall node-red-contrib-ollama from Manage palette for purpose-built nodes, or use the built-in HTTP Request node when you want fewer package dependencies.
- 3. Wire, deploy, and inspect one complete flowConfigure the reachable Ollama host, connect the prompt and response nodes, deploy the flow, and inspect the debug output before adding device or automation actions.
Best for
Builders who want visual automation flows that can call model APIs or local inference endpoints.
Model support
Node-RED ships no first-party AI nodes. Local models are reached either through the community node-red-contrib-ollama package, which wraps ollama.js with chat, generation, embedding and model-management nodes, or through a plain HTTP Request node aimed at any local API.
Choose models by role
Chat and text generation
Instruction-following Ollama model
Use a chat or instruction model for prompts, summaries, and device-facing text. Tool calling is only needed when the model itself chooses actions.
Embedding and retrieval
Qwen3 Embedding
Use a dedicated embedding model with the embedding node or direct API when a flow stores and retrieves vectors. It does not generate the final answer.
Open the model profile →How to use this integration
- Add a local model to an existing home or industrial automation flow already running on the same machine
- Wire model calls between MQTT, serial and HTTP devices without writing a service
- Generate embeddings inside a flow with the ollama-embed node for on-device search
- Prototype a model-driven flow visually on a Raspberry Pi that never contacts a cloud API
Connecting a local model
Install node-red-contrib-ollama from the palette manager and configure an ollama-config-server node, whose host defaults to localhost:11434. The Node-RED editor itself runs at http://127.0.0.1:1880. The dependency-free alternative is an http request node posting to the Ollama API directly.
Tradeoffs
This is the weakest first-party story of any runtime on this list: the Ollama nodes are a single-maintainer community contribution, not part of Node-RED, so the integration depends on one npm package keeping pace with upstream. If that matters for a long-lived deployment, the http request node has no such dependency and is the more durable choice.