Plugins

Steamship Plugins perform specific tasks related to Language AI.

Each plugin is a stateless, Python-based microservice that runs in the cloud and conforms to a strict interface and data model. Plugins may do work themselves, or they may adapt work done by third-party services for use with Steamship.

Steamship supports the following types of plugins:

• Importers
• Blockifiers
• Taggers
• Embedders

Plugins are intended to be created rarely and used prolifically. For example, one might create:

• A Notion File Importer Plugin to import a Notion page as Notion-formatted JSON

• A Notion Blockifier to convert Notion-formatted JSON into Steamship Block format

• A OpenAI Embedder to embed sentences to Vectors via GPT-3

Having created such plugins, anyone could use them to import, embed, and perform question-answering queries over their data in Notion. If someone were to create plugin for another embedding service, or another data source, it could be mixed and matched with existing plugins as well. In this way, Steamship plugins lets developers build packages across different data and AI services without worrying about the details of tasking, persistence, and integration.

Additional Links

• Developing Plugins
  • Plugin Project Structure
  • Developing Importers
  • Developing Blockifiers
    • The Blockifier Contract
    • How to Structure Blocks and Tags
    • Synchronous Example: A Pseudo-Markdown Blockifier
    • Asynchronous Blockifiers
  • Developing Taggers
  • Developing Embedders
  • Writing Async Plugins
    • Signaling an Asynchronous Response
    • Detecting an Asynchronous Progress Request
    • Throwing errors from an asynchronous plugin