Skip to main content

Ollama

Overview​

Ollama is an open-source platform that simplifies running Large Language Models (LLMs for short) on your local network.

Ollama can serve LLMs to perform various functions through which Bottle Rocket can then integrate - primarily OCR / machine vision models that allow Bottle Rocket captured scans to be interpreted as raw text data for further processing.

Install Ollama​

Pull a model​

  • Ollama allows you to pull models from the Ollama library to run on your local network. These models are improving and emerging over time; additionally, you can also train and publish your own models on Ollama.
  • Bottle Rocket is (currently) focussed on using Ollama models for OCR and text extraction from the captured images, thus, you will want to pull a model with image analysis capabilities.
    • Additionally, you will need to ensure that the model you are using is capable of handwriting recognition, as some OCR models are more suited to structured / printed text and will give limited results with scanned handwriting.

Run the pull command​

  • Inside the container or on the host machine with the Ollama installed, run the following command to 'pull' the LLM:
ollama pull <model_name>
tip

If running in a Docker container, ensure you have set up the volume mount correctly, so that the pulled models do not need to be re-downloaded on restart.

Add the Ollama connector to Bottle Rocket​

  • Login to your Bottle Rocket server.
  • On the left-hand sidebar, under My connectors, select Add connector.

Add workflow

  • Select the Ollama App Connector from the list of connectors, and click the Add connector button.

Add Ollama connector

  • For Network endpoint, enter the IP address or local DNS name of the host machine running Ollama.

    • If you are running Ollama in a Docker container, you may have to use a container-context URL such as host.docker.internal or 127.0.0.1.
    • Make sure to include the port number (Ollama defaults to port 11434).
    • Ensure that the endpoint value ends with a forward slash (/).

  • I.e. your endpoint will look something like http://host.docker.internal:11434/.

  • Save the connector by clicking Submit.