agents.planning.proper_plan_execute

Proper Plan & Execute Implementation - Full-Featured Planning with Advanced Routing.

This module provides the advanced implementation for complex planning tasks that require sophisticated routing, state management, and search integration. It builds on the clean implementation with additional capabilities.

## Key Features

• Advanced State Management: Rich state tracking with PlanExecuteState

• Search Integration: Built-in web search capabilities via duckduckgo

• Complex Routing: Sophisticated decision logic for replanning

• Execution Tracking: Detailed tracking of step execution and results

• Reusable Components: Leverages existing p_and_e models and prompts

## Architecture

Planner (SimpleAgent with Plan model)

↓

Executor (ReactAgent with tools + ExecutionResult)

↓

Routing Decision (should_continue)

├──→ Continue: back to Executor ├──→ Replan: to Replanner └──→ End: final answer

↓

Replanner (SimpleAgent with Act model)

↓

Routing Decision (route_after_replan)

├──→ Execute: back to Executor └──→ End: final answer

## Usage

### With Search Integration

from haive.agents.planning import create_plan_execute_with_search

agent = create_plan_execute_with_search(

name=”research_planner”, planner_model=”gpt-4”, executor_model=”gpt-3.5-turbo”

)

result = agent.run(

“Research the latest AI developments and create a summary report”

)

### Custom Configuration

from haive.agents.planning import create_proper_plan_execute from haive.tools import custom_api_tool, database_tool

agent = create_proper_plan_execute(

name=”data_planner”, planner_model=”gpt-4”, executor_model=”gpt-4”, tools=[custom_api_tool, database_tool], max_replanning_steps=5, allow_parallel_execution=True

)

result = agent.run(“Analyze customer data and generate insights”)

## Advanced Features

### State Processing Functions - process_planner_output: Extracts and validates plans - process_executor_output: Tracks execution results - process_replanner_output: Handles replanning decisions

### Routing Logic - should_continue: Determines next step after execution - route_after_replan: Routes after replanning decision

## When to Use

✅ Use this implementation when: - You need sophisticated planning with search - Complex routing logic is required - Detailed execution tracking is important - You need production-ready error handling

❌ Consider alternatives when: - Simple sequential planning suffices (use clean_plan_execute) - You need evidence-based planning (use ReWOO) - Tasks are very straightforward (use ReactAgent directly)

## Status: Recommended for Complex Planning Tasks

This is the go-to implementation for production systems requiring robust planning capabilities with search integration and advanced state management.

Functions

create_plan_execute_with_search([tools])	Create Plan & Execute agent with search tools.
create_proper_plan_execute([name, planner_model, ...])	Create proper Plan & Execute agent using existing p_and_e components.
process_executor_output(state, executor_result)	Process executor output and update state.
process_planner_output(state, planner_result)	Process planner output and update state.
process_replanner_output(state, replanner_result)	Process replanner output and update state.
route_after_replan(state)	Route after replanning: continue or end.
should_continue(state)	Route after execution: continue, replan, or end.

Module Contents

agents.planning.proper_plan_execute.create_plan_execute_with_search(tools=None)

Create Plan & Execute agent with search tools.

Parameters:

tools (list | None)

Return type:

haive.agents.multi.archive.enhanced_base.MultiAgentBase

agents.planning.proper_plan_execute.create_proper_plan_execute(name='ProperPlanExecute', planner_model='gpt-4o-mini', executor_model='gpt-4o-mini', replanner_model='gpt-4o-mini', tools=None)

Create proper Plan & Execute agent using existing p_and_e components.

Parameters:

• name (str) – Name for the agent system

• planner_model (str) – Model for planning agent

• executor_model (str) – Model for execution agent

• replanner_model (str) – Model for replanning agent

• tools (list | None) – Tools available to executor

Returns:

Complete Plan & Execute system

Return type:

MultiAgentBase

agents.planning.proper_plan_execute.process_executor_output(state, executor_result)

Process executor output and update state.

Parameters:

state (haive.agents.planning.p_and_e.state.PlanExecuteState)

Return type:

dict

agents.planning.proper_plan_execute.process_planner_output(state, planner_result)

Process planner output and update state.

Parameters:

state (haive.agents.planning.p_and_e.state.PlanExecuteState)

Return type:

dict

agents.planning.proper_plan_execute.process_replanner_output(state, replanner_result)

Process replanner output and update state.

Parameters:

state (haive.agents.planning.p_and_e.state.PlanExecuteState)

Return type:

dict

agents.planning.proper_plan_execute.route_after_replan(state)

Route after replanning: continue or end.

Parameters:

state (haive.agents.planning.p_and_e.state.PlanExecuteState)

Return type:

str

agents.planning.proper_plan_execute.should_continue(state)

Route after execution: continue, replan, or end.

Parameters:

state (haive.agents.planning.p_and_e.state.PlanExecuteState)

Return type:

str