re_act

ReAct (Reason + Act) pattern for intelligent agents

What it does

The re_act plugin implements the ReAct pattern:

Acknowledge (after_user_input)

Before taking any action, the agent acknowledges the request in 1-2 sentences to show it understood — planning itself is left to the main agent.

Reflect (after_tools)

After each tool execution, the agent reflects on results and decides next steps.

Quick Start

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

def search(query: str) -> str:
    """Search the web for information."""
    return f"Results for '{query}': Python is a programming language..."

agent = Agent("assistant", tools=[search], plugins=[re_act])

agent.input("Search for Python and explain what it is")
output
💭 I'll search for information about Python, then explain what it is based on the results.
[Tool: search("Python")]
💭 The search returned that Python is a programming language. I now have enough context to explain.
Python is a high-level, interpreted programming language known for its simplicity...

Want to customize? Run co copy re_act to get an editable copy.

How it works

1. Acknowledge Phase

After receiving user input, the plugin acknowledges the request (not a plan — just confirms understanding):

main.py
python
# Internal: acknowledge_request handler
@after_user_input
def acknowledge_request(agent):
    user_prompt = agent.current_session.get('user_prompt', '')
    conversation = _format_conversation(agent.current_session.get('messages', []))

    ack = llm_do(
        f"Conversation so far:\n{conversation}\n\nCurrent user input: {user_prompt}\n\nAcknowledge this request (1-2 sentences):",
        model="co/gemini-2.5-flash",
        system_prompt=ACKNOWLEDGE_PROMPT
    )

    agent.current_session['intent'] = ack
    agent.current_session['messages'].append({
        'role': 'assistant',
        'content': ack
    })

2. Reflection Phase

After each batch of tools executes, the plugin reflects on the most recent result:

main.py
python
# Uses the built-in reflect handler from useful_events_handlers
@after_tools
def reflect(agent):
    trace = agent.current_session['trace'][-1]

    if trace['type'] != 'tool_result':
        return

    tool_name = trace['name']
    tool_args = trace['args']

    if trace['status'] == 'success':
        reflection = llm_do(
            f"Action: {tool_name}({tool_args})\nResult: {str(trace['result'])[:300]}\nWhat did we learn? What's next?",
            model="co/gemini-2.5-flash"
        )
        agent.current_session['messages'].append({
            'role': 'assistant',
            'content': reflection
        })

Combined with Eval Plugin

For debugging and testing, combine with the eval plugin:

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

# ReAct provides planning, eval provides evaluation at completion
agent = Agent("assistant", tools=[search], plugins=[re_act, eval])

agent.input("Search for Python")
# Output includes:
# Acknowledge: "I'll search for Python information and explain it to you."
# [Tool execution]
# Reflection: We learned Python is a programming language
# ✓ Evaluation: Task complete - found and understood Python info

Events Used

Event	Handler	Purpose
`after_user_input`	acknowledge_request	Acknowledge what the user is asking for
`after_tools`	reflect	Reflect on tool results

Source

connectonion/useful_plugins/re_act.py

main.py
python
# The plugin is just a list of event handlers
re_act = [acknowledge_request, reflect]

Useful Plugins

eval