A GitHub action to sync files from your repository to an OpenAI Assistants v2 Vector Store.

Perfect for turning a static website, Obsidian vault, or a project's source code into a knowledge base for a ChatGPT assistant.

• Only uploads changes on each run.

• Will remove files from the vector store that are removed from the repository.

• Can differentiate from other Files and Vector Stores within the organization.

• Will filter out both unsuported and empty files.

• Defaults to all markdown files in the repository, but can be set to other files types or paths by using a glob pattern.

• pattern: The glob pattern matching the files to sync. Default: **/*.md.
• token: The OpenAI API token. Default: ${{ secrets.OPENAI_API_KEY }}.
• key: A key used to identify the synced files and the vector store. Defaults to the repository name.

- uses: actions/checkout@v3

- name: Sync markdown files to Vector Store
  uses: shmatt/assistants-vector-store-sync@v1
  with:
    pattern: "**/*.md"
    token: ${{ secrets.OPENAI_API_KEY }}
    key: knowledge-base

The action uses a key, which is derived from the repository name (without the owner prefix) but may be provided via the key input.

It will look for a Vector Store that has metadata with the metadata key key set to the provided value / repository name. If one doesn't exist, it will be created with the name also set to the value.

For example, if the key or repostory name is knowledge-base, the Vector Store as JSON would look like:

{
  ...
  "name": "knowledge-base",
  "metatadata": {
    "key": "knowledge-base"
    ...
  }
}

The relative paths of matching files from the repository are prefixed with {key}-{md5hash}/, with the md5hash being a hash of the file contents. This key / hash combination is to enable the action to track the files. The name of the uploaded file is then set to the result.

The action will then;

1. upload new or changed files;
2. remove deleted or replaced files; and
3. ensure files are added to the Vector Store.

You can then attach the Vector Store to an Assistant using the tool_resources field in your code, or manually in the OpenAI Assistants playground.

Contributions are welcome! Please open an issue or a pull request.