DocsUseful Plugins

Useful Plugins

Built-in

Pre-built plugins from connectonion.useful_plugins

What is a Plugin?

A plugin is a reusable list of event handlers. Use plugins=[...] to add pre-packaged functionality to any agent.

main.py
python
from connectonion import Agent
from connectonion.useful_plugins import re_act, image_result_formatter

# Combine multiple plugins
agent = Agent(
    "assistant",
    tools=[search, screenshot],
    plugins=[re_act, image_result_formatter]
)

Learn how to build custom plugins in the Plugin System documentation.

ReAct Pattern

re_act

Acknowledges the user's request before acting, then reflects on what was learned after each batch of tool calls

When your agent needs to reason step-by-step before acting. Improves accuracy on multi-step tasks.

zero-configafter_user_input (acknowledge)after_tools (reflect)

Task Evaluation

eval

Generates an expected outcome before the task runs, then LLM-judges whether the agent actually achieved it

When you want automated pass/fail scoring of whether your agent completed the task correctly.

zero-configafter_user_input (generate_expected)on_complete (evaluate_completion)

Image Result Formatter

image_result_formatter

Automatically formats base64 images in tool results for vision models (GPT-4o, etc.)

When a tool returns a screenshot or image and you want the vision model to actually see it.

zero-configafter_tools

Gmail Plugin

gmail_plugin

Pre-configured event handlers for Gmail integration workflows

When building an email assistant. Handles read/reply/label workflows out of the box.

needs Gmail authbefore_each_tool (approval)after_each_tool (CRM sync)

Calendar Plugin

calendar_plugin

Pre-configured event handlers for calendar management workflows

When building a scheduling assistant. Handles event creation and availability checks.

needs Google authbefore_each_tool

Shell Approval

shell_approval

Requires user confirmation before executing shell commands for safety

Essential when your agent runs shell commands in production. Prevents accidental destructive ops.

zero-configbefore_each_tool

System Reminder

system_reminder

Injects contextual guidance into tool results to nudge agent behavior

When the agent keeps making the same mistake. Inject a correction reminder after each tool call.

zero-configafter_each_tool

Combining Plugins

Plugins can be combined for powerful agent behaviors:

main.py
python
from connectonion import Agent
from connectonion.useful_plugins import (
    re_act,              # Planning + Reflection
    image_result_formatter,  # Vision support
    shell_approval       # Safety for shell commands
)

agent = Agent(
    "devops_assistant",
    tools=[run_command, screenshot, analyze],
    plugins=[re_act, image_result_formatter, shell_approval]
)

# Now the agent will:
# 1. Plan before acting (re_act)
# 2. Format images for vision models (image_result_formatter)
# 3. Ask for approval before shell commands (shell_approval)
# 4. Reflect after tool execution (re_act)

Customizing Plugins

Need to modify a built-in plugin? Copy it to your project:

code
bash
# Copy a plugin to your project
co copy re_act

# Copy multiple plugins
co copy re_act shell_approval eval
output
✓ Copied: ./plugins/re_act.py
✓ Copied: ./plugins/shell_approval.py
✓ Copied: ./plugins/eval.py

Then import from your local copy and customize:

main.py
python
# Before (from package)
from connectonion.useful_plugins import re_act

# After (from your copy)
from plugins.re_act import re_act  # Customize freely!