Plugins & Plugin Packages

This document provides reference information for plugins and plugin packages in FoundationaLLM.

Overview

Plugins extend the functionality of the FoundationaLLM platform. Plugins can be used to add new features, integrate with external systems, or perform other tasks. Plugins can be managed using the FoundationaLLM Management Portal (interactively) or the FoundationaLLM Management API (programmatically).

Plugin Packages

A FoundationaLLM plugin package is a unit of versioning and deployment of one or more plugins.

Supported Package Types

Platform	Package Type	Description
Python	ZIP package	Contains agent workflow and agent tool plugins. Scheduled to be replaced by Wheel packages.
.NET	NuGet package	Contains data source, data pipeline stage, content text extraction, and content text partitioning plugins. Fully supported by Management API.

Package Naming Convention

Plugin package names must follow a strict naming convention: {platform}-{name}, where:

{platform} can be Dotnet or Python
{name} can only contain alphanumerical characters, underlines, or hyphens

Example: Dotnet-FoundationaLLMDataPipelinePlugins

Managing Plugin Packages via API

Create or Update a Plugin Package

POST /instances/{instanceId}/providers/FoundationaLLM.Plugin/pluginPackages/{packageName}

Request body:

Must be of type form-data
Must contain a file with the key file (the NuGet package)
Must contain a text with key resource:

{
    "type": "plugin-package",
    "name": "{packageName}"
}

List Plugin Packages

GET /instances/{instanceId}/providers/FoundationaLLM.Plugin/pluginPackages

Plugins

Plugin Types

Platform	Category	Description
Python	Agent Workflow	Implements agent workflows running in LangChain API
Python	Agent Tool	Implements agent tools for LangChain workflow host
.NET	Data Source	Implements data sources for data pipelines
.NET	Data Pipeline Stage	Implements stages for data pipeline execution
.NET	Content Text Extraction	Implements text extraction from binary content
.NET	Content Text Partitioning	Implements text partitioning strategies

Plugin Naming Convention

Plugin names must follow the format {platform}-{packageName}-{pluginName}, where:

{platform} can be Dotnet or Python
{packageName} is the plugin package name (alphanumerical, underlines, hyphens only)
{pluginName} is the plugin name (alphanumerical, underlines, hyphens only)

Example: Dotnet-FoundationaLLMDataPipelinePlugins-AzureAISearchIndexingDataPipelineStage

Plugin Definition Structure

{
    "type": "plugin",
    "name": "Dotnet-FoundationaLLMDataPipelinePlugins-AzureAISearchIndexingDataPipelineStage",
    "object_id": "instances/{instanceId}/providers/FoundationaLLM.Plugin/plugins/...",
    "display_name": "Azure AI Search Indexing Data Pipeline Stage (FoundationaLLM)",
    "description": "Provides the FoundationaLLM standard implementation for indexing data pipeline stages.",
    "category": "Data Pipeline Stage",
    "parameters": [...],
    "parameter_selection_hints": {...},
    "dependencies": [...]
}

Plugin Parameters

Each plugin has zero or more parameters. Supported parameter types:

Type	Description
`string`	A single string value
`int`	A single integer value
`float`	A single floating-point value
`bool`	A single boolean value
`datetime`	A single date and time value
`array`	An array of values
`resource-object-id`	A FoundationaLLM resource identifier

Parameter Selection Hints

For resource-object-id parameters, the parameter_selection_hints property provides UI guidance:

"parameter_selection_hints": {
    "APIEndpointConfigurationObjectId": {
        "resourcePath": "providers/FoundationaLLM.Configuration/apiEndpointConfigurations",
        "filterActionPayload": {
            "Category": "General",
            "Subcategory": "Indexing"
        }
    }
}

Plugin Dependencies

Plugins can have dependencies on other plugins. Dependency types:

Single: Exactly one dependency plugin must be selected
Multiple: One or more dependency plugins must be selected

"dependencies": [
    {
        "selection_type": "Single",
        "dependency_plugin_names": [
            "Dotnet-FoundationaLLMDataPipelinePlugins-TokenContentTextPartitioning",
            "Dotnet-FoundationaLLMDataPipelinePlugins-SemanticContentTextPartitioning"
        ]
    }
]

Managing Plugins via API

List All Plugins

GET /instances/{instanceId}/providers/FoundationaLLM.Plugin/plugins

Filter Plugins by Category

POST /instances/{instanceId}/providers/FoundationaLLM.Plugin/plugins/filter