Plugin

EvoFabric supports user-defined plugins, and the currently predefined plugin types are:

• SyncNode: Sync node

• AsyncNode: asynchronous node

• SyncStreamNode: Sync Stream Node

• AsyncStreamNode: Async Stream Node

• ChatClientBase: Custom Large Model

• EmbedClientBase: Custom text Embedding module

• RerankClientBase: Custom ReRank module

• MemBase: Custom Memory Module

• ToolManager: Custom Tool Manager

• McpToolManager: MCP Tool Manager

• CodeSandbox: Code Sandbox

• DBBase: Database

Note

Supported plugin types can also be looked up in evofabric.plugin_manager.PluginTypeDict

Plugin Integration Principle

Plugins integrating with the EvoFabric framework require the use of Python’s entry-points mechanism: when users define entry-points in the configuration file of their plugin package, EvoFabric will automatically recognize the plugin.

Example as follows:

# ChatClientBase type plugin
[project.entry-points."ChatClientBase"]
# 'demo_tool1' is plugin name
demo_tool1 = "demo_tool1:create"

The above configuration demo_tool1:create() corresponds to the demo_tool1.create function, and EvoFabric will call this function during the plugin registration phase to obtain the plugin class.

Plugin initialization logic is visible: evofabric.plugin_manager.load_plugins()

After being loaded by the EvoFabric framework, plugins automatically inherit from the parent class corresponding to their type, and users can then use the corresponding plugins as needed.

Plugin Example

Next, we can customize a plugin named demo_tool1 according to this guide:

1. First, we need to create a new Python package. The directory structure is as follows:

demo_tool1
│  pyproject.toml
│  README.md
│
├─demo_tool1
│  │ invoke_tool.py
│  │ __init__.py

The pyproject.toml file contains the entry-points registration block, the specific configuration is as follows:

# pyproject.toml

[build-system]
requires = ["setuptools>=61.0"]
build-backend = "setuptools.build_meta"

[project]
name = "demo_tool1"
version = "0.1.3"
authors = [
  { name="author", email="author@example.com" },
]
description = "An Agent System plugin demo."
readme = "README.md"
requires-python = ">=3.11"
classifiers = [
    "Programming Language :: Python :: 3",
    "License :: OSI Approved :: MIT License",
    "Operating System :: OS Independent",
]

dependencies = []

[project.entry-points."ChatClientBase"]
# 'demo_tool1' is plugin name.
demo_tool1 = "demo_tool1:register"

2. Write plugin logic. The following is a streaming output example of an LLM module, located at demo_tool1/invoke_tool.py:

from pydantic import Field


class LLMRunner:
state: str = Field(description="zhuan")

async def create_on_stream(
        self, messages
):
    for token in "This is the implementation of create_on_stream.":
        yield token

3. Add registration function, located at demo_tool1/invoke_tool.py:

def register():
    return "demo_tool1.LLMRunner"

The plugin has been completed. Simply execute pip install -e . in the current directory to install it.