"""
Pattern 028: Working Memory
Description:
The Working Memory pattern provides an active workspace for the agent's current
task processing. Unlike short-term memory (conversation history) or long-term
memory (persistent storage), working memory holds intermediate results, active
goals, and context needed for the current cognitive task. It's dynamically
updated as the agent progresses through multi-step reasoning.
Components:
- Active Goals: Current objectives being pursued
- Intermediate Results: Partial solutions and computations
- Task Context: Relevant information for current task
- Attention Focus: What the agent is currently processing
- Scratch Space: Temporary storage for calculations
Use Cases:
- Multi-step problem solving
- Complex reasoning tasks
- Task decomposition and execution
- Mathematical computations
- Planning and execution monitoring
LangChain Implementation:
Uses structured state management to track active goals, intermediate
results, and task progress. Implements focus mechanisms to maintain
relevant context during complex operations.
Key Features:
- Task-scoped memory
- Goal tracking
- Intermediate result storage
- Dynamic updates
- Context relevance filtering
"""
import os
from typing import List, Dict, Any, Optional, Set
from dataclasses import dataclass, field
from enum import Enum
from datetime import datetime
from dotenv import load_dotenv
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_openai import ChatOpenAI
load_dotenv()
class GoalStatus(Enum):
"""Status of a goal in working memory."""
ACTIVE = "active"
IN_PROGRESS = "in_progress"
COMPLETED = "completed"
BLOCKED = "blocked"
ABANDONED = "abandoned"
@dataclass
class Goal:
"""Represents a goal in working memory."""
id: str
description: str
status: GoalStatus = GoalStatus.ACTIVE
parent_goal_id: Optional[str] = None
sub_goals: List[str] = field(default_factory=list)
result: Optional[Any] = None
metadata: Dict[str, Any] = field(default_factory=dict)
created_at: datetime = field(default_factory=datetime.now)
def to_dict(self) -> Dict[str, Any]:
"""Convert to dictionary."""
return {
"id": self.id,
"description": self.description,
"status": self.status.value,
"parent_goal_id": self.parent_goal_id,
"sub_goals": self.sub_goals,
"result": str(self.result) if self.result is not None else None,
"metadata": self.metadata
}
@dataclass
class IntermediateResult:
"""Stores intermediate computation results."""
id: str
description: str
value: Any
goal_id: str
timestamp: datetime = field(default_factory=datetime.now)
def to_dict(self) -> Dict[str, Any]:
"""Convert to dictionary."""
return {
"id": self.id,
"description": self.description,
"value": str(self.value),
"goal_id": self.goal_id,
"timestamp": self.timestamp.isoformat()
}
class WorkingMemorySpace:
"""
Working memory workspace for active task processing.
"""
def __init__(self, capacity: int = 7):
"""
Initialize working memory.
Args:
capacity: Maximum number of active items (Miller's Law: 7±2)
"""
self.capacity = capacity
self.goals: Dict[str, Goal] = {}
self.intermediate_results: Dict[str, IntermediateResult] = {}
self.task_context: Dict[str, Any] = {}
self.attention_focus: Optional[str] = None # Current goal ID
self.scratch_space: Dict[str, Any] = {}
def add_goal(
self,
goal_id: str,
description: str,
parent_goal_id: Optional[str] = None
) -> Goal:
"""
Add a new goal to working memory.
Args:
goal_id: Unique goal identifier
description: Goal description
parent_goal_id: Parent goal if this is a sub-goal
Returns:
Created goal
"""
goal = Goal(
id=goal_id,
description=description,
parent_goal_id=parent_goal_id
)
self.goals[goal_id] = goal
# Link to parent
if parent_goal_id and parent_goal_id in self.goals:
self.goals[parent_goal_id].sub_goals.append(goal_id)
# Set as focus if first goal
if self.attention_focus is None:
self.attention_focus = goal_id
return goal
def update_goal_status(self, goal_id: str, status: GoalStatus, result: Any = None):
"""Update goal status and optionally store result."""
if goal_id in self.goals:
self.goals[goal_id].status = status
if result is not None:
self.goals[goal_id].result = result
def store_intermediate(
self,
result_id: str,
description: str,
value: Any,
goal_id: str
):
"""Store an intermediate result."""
result = IntermediateResult(
id=result_id,
description=description,
value=value,
goal_id=goal_id
)
self.intermediate_results[result_id] = result
def get_active_goals(self) -> List[Goal]:
"""Get all active goals."""
return [
goal for goal in self.goals.values()
if goal.status in [GoalStatus.ACTIVE, GoalStatus.IN_PROGRESS]
]
def get_current_focus(self) -> Optional[Goal]:
"""Get currently focused goal."""
if self.attention_focus and self.attention_focus in self.goals:
return self.goals[self.attention_focus]
return None
def shift_focus(self, goal_id: str):
"""Shift attention to a different goal."""
if goal_id in self.goals:
self.attention_focus = goal_id
def get_goal_context(self, goal_id: str) -> Dict[str, Any]:
"""Get all relevant context for a goal."""
if goal_id not in self.goals:
return {}
goal = self.goals[goal_id]
context = {
"goal": goal.to_dict(),
"intermediate_results": [
ir.to_dict() for ir in self.intermediate_results.values()
if ir.goal_id == goal_id
],
"sub_goals": [
self.goals[sg_id].to_dict()
for sg_id in goal.sub_goals
if sg_id in self.goals
]
}
return context
def get_capacity_usage(self) -> float:
"""Get percentage of working memory capacity used."""
active_items = len(self.get_active_goals()) + len(self.intermediate_results)
return min(100.0, (active_items / self.capacity) * 100)
def clear(self):
"""Clear working memory."""
self.goals.clear()
self.intermediate_results.clear()
self.task_context.clear()
self.attention_focus = None
self.scratch_space.clear()
def get_memory_state(self) -> Dict[str, Any]:
"""Get complete state of working memory."""
return {
"capacity": self.capacity,
"capacity_usage": self.get_capacity_usage(),
"total_goals": len(self.goals),
"active_goals": len(self.get_active_goals()),
"intermediate_results": len(self.intermediate_results),
"current_focus": self.attention_focus,
"goals": [g.to_dict() for g in self.goals.values()],
"results": [ir.to_dict() for ir in self.intermediate_results.values()]
}
class WorkingMemoryAgent:
"""
Agent with working memory for complex task processing.
"""
def __init__(self, model: str = "gpt-3.5-turbo"):
"""
Initialize agent with working memory.
Args:
model: LLM model to use
"""
self.llm = ChatOpenAI(model=model, temperature=0.3)
self.working_memory = WorkingMemorySpace()
self.task_counter = 0
def solve_problem(self, problem: str) -> Dict[str, Any]:
"""
Solve a problem using working memory.
Args:
problem: Problem description
Returns:
Solution with working memory trace
"""
print(f"\n[Agent] Solving problem: {problem}")
print(f"[Agent] Initializing working memory...\n")
# Clear working memory for new task
self.working_memory.clear()
# Create main goal
main_goal_id = "goal_0"
self.working_memory.add_goal(main_goal_id, problem)
# Decompose problem into sub-goals
print("[Step 1] Decomposing problem into sub-goals...")
sub_goals = self._decompose_problem(problem)
for i, sub_goal in enumerate(sub_goals):
sub_goal_id = f"goal_{i+1}"
self.working_memory.add_goal(sub_goal_id, sub_goal, main_goal_id)
print(f" Sub-goal {i+1}: {sub_goal}")
print(f"\n[Working Memory] Capacity: {self.working_memory.get_capacity_usage():.1f}%\n")
# Process each sub-goal
for i, sub_goal in enumerate(sub_goals):
sub_goal_id = f"goal_{i+1}"
# Shift focus to this sub-goal
self.working_memory.shift_focus(sub_goal_id)
current_focus = self.working_memory.get_current_focus()
print(f"[Step {i+2}] Processing: {current_focus.description}")
# Mark as in progress
self.working_memory.update_goal_status(sub_goal_id, GoalStatus.IN_PROGRESS)
# Solve sub-goal
result = self._solve_sub_goal(sub_goal)
# Store result
self.working_memory.update_goal_status(
sub_goal_id,
GoalStatus.COMPLETED,
result
)
# Store as intermediate result
self.working_memory.store_intermediate(
f"result_{i}",
f"Solution to: {sub_goal}",
result,
sub_goal_id
)
print(f" ✓ Result: {result}")
print(f" [Working Memory] Capacity: {self.working_memory.get_capacity_usage():.1f}%\n")
# Synthesize final solution
print("[Final Step] Synthesizing solution from sub-goals...")
final_solution = self._synthesize_solution(problem)
# Mark main goal as completed
self.working_memory.update_goal_status(
main_goal_id,
GoalStatus.COMPLETED,
final_solution
)
print(f"\n✓ Solution: {final_solution}\n")
# Get memory state
memory_state = self.working_memory.get_memory_state()
return {
"problem": problem,
"solution": final_solution,
"sub_goals": sub_goals,
"working_memory_trace": memory_state
}
def _decompose_problem(self, problem: str) -> List[str]:
"""Decompose problem into sub-goals."""
prompt = ChatPromptTemplate.from_messages([
("system", """You are an expert at problem decomposition. Break down
the given problem into 3-5 clear, actionable sub-goals."""),
("user", """Problem: {problem}
List 3-5 sub-goals needed to solve this problem. Be specific and actionable.
Respond with one sub-goal per line.""")
])
chain = prompt | self.llm | StrOutputParser()
response = chain.invoke({"problem": problem})
# Parse sub-goals
lines = response.strip().split("\n")
sub_goals = []
for line in lines:
line = line.strip()
# Remove numbering
for prefix in ["1.", "2.", "3.", "4.", "5.", "-", "*", "•"]:
if line.startswith(prefix):
line = line[len(prefix):].strip()
if line:
sub_goals.append(line)
return sub_goals[:5] # Limit to 5 sub-goals
def _solve_sub_goal(self, sub_goal: str) -> str:
"""Solve a single sub-goal."""
# Get context from working memory
focus = self.working_memory.get_current_focus()
context = self.working_memory.get_goal_context(focus.id) if focus else {}
# Get previously completed results
previous_results = [
ir for ir in self.working_memory.intermediate_results.values()
]
previous_context = ""
if previous_results:
previous_context = "\n".join([
f"- {ir.description}: {ir.value}"
for ir in previous_results
])
prompt = ChatPromptTemplate.from_messages([
("system", "You solve sub-goals clearly and concisely."),
("user", """Sub-goal: {sub_goal}
Previous Results:
{previous_context}
Provide a clear, concise solution:""")
])
chain = prompt | self.llm | StrOutputParser()
result = chain.invoke({
"sub_goal": sub_goal,
"previous_context": previous_context if previous_context else "None yet"
})
return result.strip()
def _synthesize_solution(self, problem: str) -> str:
"""Synthesize final solution from sub-goal results."""
# Get all intermediate results
results_text = "\n".join([
f"{i+1}. {ir.description}\n {ir.value}"
for i, ir in enumerate(self.working_memory.intermediate_results.values())
])
prompt = ChatPromptTemplate.from_messages([
("system", "You synthesize complete solutions from sub-goal results."),
("user", """Original Problem: {problem}
Sub-goal Results:
{results_text}
Provide a complete, coherent solution to the original problem:""")
])
chain = prompt | self.llm | StrOutputParser()
solution = chain.invoke({
"problem": problem,
"results_text": results_text
})
return solution.strip()
def demonstrate_working_memory():
"""Demonstrate the Working Memory pattern."""
print("=" * 80)
print("WORKING MEMORY PATTERN DEMONSTRATION")
print("=" * 80)
agent = WorkingMemoryAgent()
# Test 1: Multi-step math problem
print("\n" + "=" * 80)
print("TEST 1: Multi-Step Mathematical Problem")
print("=" * 80)
problem1 = """A store sells apples for $2 each and oranges for $3 each.
If you buy 5 apples and 4 oranges, get a 10% discount, and pay 8% tax,
what is the final price?"""
result1 = agent.solve_problem(problem1)
print("=" * 80)
print("WORKING MEMORY TRACE")
print("=" * 80)
print(f"Goals tracked: {result1['working_memory_trace']['total_goals']}")
print(f"Active computations: {result1['working_memory_trace']['intermediate_results']}")
print(f"Capacity used: {result1['working_memory_trace']['capacity_usage']:.1f}%")
# Test 2: Planning task
print("\n\n" + "=" * 80)
print("TEST 2: Planning Task")
print("=" * 80)
problem2 = """Plan a weekend trip to a nearby city including:
transportation, accommodation, activities, and budget estimation."""
result2 = agent.solve_problem(problem2)
print("=" * 80)
print("WORKING MEMORY TRACE")
print("=" * 80)
print(f"Goals tracked: {result2['working_memory_trace']['total_goals']}")
print(f"Sub-goals completed: {len(result2['sub_goals'])}")
# Test 3: Analytical task
print("\n\n" + "=" * 80)
print("TEST 3: Analytical Reasoning")
print("=" * 80)
problem3 = """Analyze the pros and cons of remote work and provide
a recommendation for a tech company."""
result3 = agent.solve_problem(problem3)
print("=" * 80)
print("WORKING MEMORY TRACE")
print("=" * 80)
for goal in result3['working_memory_trace']['goals']:
print(f"\nGoal: {goal['description']}")
print(f" Status: {goal['status']}")
if goal['result']:
preview = goal['result'][:100] + "..." if len(goal['result']) > 100 else goal['result']
print(f" Result: {preview}")
# Summary
print("\n\n" + "=" * 80)
print("PATTERN SUMMARY")
print("=" * 80)
print("""
The Working Memory pattern demonstrates:
1. **Active Workspace**: Maintains current task context
2. **Goal Tracking**: Monitors progress on objectives
3. **Intermediate Storage**: Stores partial results
4. **Attention Focus**: Tracks current processing target
5. **Capacity Management**: Limited capacity (Miller's Law)
Key Components:
Goals:
- Hierarchical goal structure (main goal → sub-goals)
- Status tracking (active, in-progress, completed, blocked)
- Result storage
- Parent-child relationships
Intermediate Results:
- Temporary computation storage
- Linked to specific goals
- Time-stamped for ordering
- Available for later steps
Attention Focus:
- Single goal actively processed
- Context switching between goals
- Maintains processing coherence
Working Memory Characteristics:
Capacity Limitation (Miller's Law):
- 7±2 items simultaneously
- Prevents cognitive overload
- Forces prioritization
- Matches human working memory
Dynamic Updates:
- Content changes as task progresses
- Old items removed when complete
- New items added as needed
- Maintains relevance
Task-Scoped:
- Cleared between tasks
- Fresh workspace for each problem
- No cross-task contamination
Differences from Other Memory Types:
vs Short-Term Memory:
- Working: Active processing workspace
- Short-term: Conversation history buffer
- Working is goal-oriented, short-term is sequential
vs Long-Term Memory:
- Working: Temporary, task-scoped
- Long-term: Permanent, cross-session
- Working is volatile, long-term persists
Use Cases:
- Multi-step problem solving
- Complex calculations
- Task planning and execution
- Reasoning chains
- Resource allocation
Benefits:
- **Organized Processing**: Structured approach to complexity
- **Progress Tracking**: Monitor goal completion
- **Context Maintenance**: Relevant information available
- **Intermediate Access**: Use partial results
- **Cognitive Modeling**: Mimics human working memory
This pattern is essential for agents tackling complex, multi-step
tasks that require maintaining and manipulating intermediate state.
""")
if __name__ == "__main__":
demonstrate_working_memory()
Agentic Design Pattern: Working Memory
Pattern 028: Working Memory – Line-by-Line Explanation
Overview
This document provides a detailed line-by-line explanation of the Working Memory pattern implementation.
Pattern Description
Pattern 028: Working Memory
Description: The Working Memory pattern provides an active workspace for the agent’s current task processing. Unlike short-term memory (conversation history) or long-term memory (persistent storage), working memory holds intermediate results, active goals, and context needed for the current cognitive task. It’s dynamically updated as the agent progresses through multi-step reasoning.
Components: – Active Goals: Current objectives being pursued – Intermediate Results: Partial solutions and computations – Task Context: Relevant information for current task – Attention Focus: What the agent is currently processing – Scratch Space: Temporary storage for calculations
Use Cases: – Multi-step problem solving – Complex reasoning tasks – Task decomposition and execution – Mathematical computations – Planning and execution monitoring
LangChain Implementation: Uses structured state management to track active goals, intermediate results, and task progress. Implements focus mechanisms to maintain relevant context during complex operations.
Key Features: – Task-scoped memory – Goal tracking – Intermediate result storage – Dynamic updates – Context relevance filtering
File Statistics
- Total Lines: 594
- Classes: 5
- Functions: 1
- Constants: 0
Module Documentation
"""
Pattern 028: Working Memory
Description:
The Working Memory pattern provides an active workspace for the agent's current
task processing. Unlike short-term memory (conversation history) or long-term
memory (persistent storage), working memory holds intermediate results, active
goals, and context needed for the current cognitive task. It's dynamically
updated as the agent progresses through multi-step reasoning.
Components:
- Active Goals: Current objectives being pursued
- Intermediate Results: Partial solutions and computations
- Task Context: Relevant information for current task
- Attention Focus: What the agent is currently processing
- Scratch Space: Temporary storage for calculations
Use Cases:
- Multi-step problem solving
- Complex reasoning tasks
- Task decomposition and execution
- Mathematical computations
- Planning and execution monitoring
LangChain Implementation:
Uses structured state management to track active goals, intermediate
results, and task progress. Implements focus mechanisms to maintain
relevant context during complex operations.
Key Features:
- Task-scoped memory
- Goal tracking
- Intermediate result storage
- Dynamic updates
- Context relevance filtering
"""
Purpose: This module implements the Working Memory pattern, providing a structured approach to agent design and implementation.
Import Statements
Standard Library Imports
import dataclasses.dataclass
import dataclasses.field
import datetime.datetime
import enum.Enum
import langchain_core.output_parsers.StrOutputParser
import langchain_core.prompts.ChatPromptTemplate
import os
import typing.Any
import typing.Dict
import typing.List
import typing.Optional
import typing.Set
Purpose: Core Python functionality:
– dataclasses.dataclass: Simplified class definitions with @dataclass
– dataclasses.field: Simplified class definitions with @dataclass
– datetime.datetime: Date and time handling
– enum.Enum: Enumeration support
– langchain_core.output_parsers.StrOutputParser: Regular expression operations
– langchain_core.prompts.ChatPromptTemplate: Regular expression operations
– os: Environment variable access and system operations
– typing.Any: Type hints for better code documentation (List, Dict, Any, Optional)
– typing.Dict: Type hints for better code documentation (List, Dict, Any, Optional)
– typing.List: Type hints for better code documentation (List, Dict, Any, Optional)
– typing.Optional: Type hints for better code documentation (List, Dict, Any, Optional)
– typing.Set: Type hints for better code documentation (List, Dict, Any, Optional)
LangChain Imports
from langchain_openai.ChatOpenAI
Purpose: LangChain framework components: – LLM Interface: Wrapper for OpenAI models – Chains: Composable building blocks for complex workflows
Other Imports
from dotenv.load_dotenv
Purpose: Additional dependencies for specialized functionality.
Classes
Class: GoalStatus
Lines 52-58
Inherits from: Enum
Purpose:
Status of a goal in working memory.
Class: Goal
Lines 62-83
Purpose:
Represents a goal in working memory.
Methods:
to_dict(self)
Lines 73-83
Purpose:
Convert to dictionary.
Parameters:
Returns: <ast.Subscript object at 0x000001A06D66AB90>
Class: IntermediateResult
Lines 87-103
Purpose:
Stores intermediate computation results.
Methods:
to_dict(self)
Lines 95-103
Purpose:
Convert to dictionary.
Parameters:
Returns: <ast.Subscript object at 0x000001A06D6A3390>
Class: WorkingMemorySpace
Lines 106-246
Purpose:
Working memory workspace for active task processing.
Methods:
__init__(self, capacity)
Lines 111-123
Purpose:
Initialize working memory.
Args:
capacity: Maximum number of active items (Miller's Law: 7±2)
Parameters:
– capacity: Capacity
add_goal(self, goal_id, description, parent_goal_id)
Lines 125-158
Purpose:
Add a new goal to working memory.
Args:
goal_id: Unique goal identifier
description: Goal description
parent_goal_id: Parent goal if this is a sub-goal
Returns:
Created goal
Parameters:
– goal_id: Goal id
– description: Description
– parent_goal_id: Parent goal id
Returns: Goal
update_goal_status(self, goal_id, status, result)
Lines 160-165
Purpose:
Update goal status and optionally store result.
Parameters:
– goal_id: Goal id
– status: Status
– result: Result
store_intermediate(self, result_id, description, value, goal_id)
Lines 167-181
Purpose:
Store an intermediate result.
Parameters:
– result_id: Result id
– description: Description
– value: Value
– goal_id: Goal id
get_active_goals(self)
Lines 183-188
Purpose:
Get all active goals.
Parameters:
Returns: <ast.Subscript object at 0x000001A06D156A50>
get_current_focus(self)
Lines 190-194
Purpose:
Get currently focused goal.
Parameters:
Returns: <ast.Subscript object at 0x000001A06D658710>
shift_focus(self, goal_id)
Lines 196-199
Purpose:
Shift attention to a different goal.
Parameters:
– goal_id: Goal id
get_goal_context(self, goal_id)
Lines 201-220
Purpose:
Get all relevant context for a goal.
Parameters:
– goal_id: Goal id
Returns: <ast.Subscript object at 0x000001A06D739FD0>
get_capacity_usage(self)
Lines 222-225
Purpose:
Get percentage of working memory capacity used.
Parameters:
Returns: float
clear(self)
Lines 227-233
Purpose:
Clear working memory.
Parameters:
get_memory_state(self)
Lines 235-246
Purpose:
Get complete state of working memory.
Parameters:
Returns: <ast.Subscript object at 0x000001A06D62ABD0>
Class: WorkingMemoryAgent
Lines 249-444
Purpose:
Agent with working memory for complex task processing.
Methods:
__init__(self, model)
Lines 254-263
Purpose:
Initialize agent with working memory.
Args:
model: LLM model to use
Parameters:
– model: Language model instance
solve_problem(self, problem)
Lines 265-351
Purpose:
Solve a problem using working memory.
Args:
problem: Problem description
Returns:
Solution with working memory trace
Parameters:
– problem: Problem
Returns: <ast.Subscript object at 0x000001A06D60A890>
_decompose_problem(self, problem)
Lines 353-380
Purpose:
Decompose problem into sub-goals.
Parameters:
– problem: Problem
Returns: <ast.Subscript object at 0x000001A06D5F7C50>
_solve_sub_goal(self, sub_goal)
Lines 382-417
Purpose:
Solve a single sub-goal.
Parameters:
– sub_goal: Sub goal
Returns: str
_synthesize_solution(self, problem)
Lines 419-444
Purpose:
Synthesize final solution from sub-goal results.
Parameters:
– problem: Problem
Returns: str
Functions
demonstrate_working_memory()
Lines 447-589
Purpose:
Demonstrate the Working Memory pattern.
Key Concepts Summary
Pattern Overview
The Working Memory pattern provides a structured approach to building AI agents with specific capabilities and behaviors. This implementation leverages LangChain’s composable components to create a robust and maintainable solution.
Core Components
- Agent Class: Main implementation of the pattern logic
- Prompt Engineering: Carefully crafted prompts to guide LLM behavior
- Tool Integration: Optional tools for extended functionality
- Memory Management: State handling across interactions
- Output Parsing: Structured response processing
When to Use
✅ Good for: – Scenarios requiring this specific pattern’s strengths – Applications needing structured agent behavior – Systems where explainability is important – Production environments requiring reliability
❌ Not ideal for: – Simple single-prompt tasks – Scenarios requiring different pattern characteristics – Resource-constrained environments (if computationally intensive)
Best Practices Demonstrated
1. Environment Configuration
- ✅ Use
.envfiles for API keys - ✅ Never hardcode credentials
- ✅ Validate environment variables on startup
2. Error Handling
- ✅ Graceful degradation on failures
- ✅ Informative error messages
- ✅ Logging for debugging
3. Prompt Engineering
- ✅ Clear, specific instructions
- ✅ Structured output formats
- ✅ Few-shot examples when beneficial
4. Code Organization
- ✅ Separation of concerns (class-based design)
- ✅ Type hints for clarity
- ✅ Docstrings for documentation
5. Performance Optimization
- ✅ Appropriate temperature settings
- ✅ Token budget management
- ✅ Caching where applicable
LangChain Components Used
- ChatOpenAI: OpenAI model interface
- Chains: Sequential component composition
Why LangChain? – Composable abstractions for LLM applications – Rich ecosystem of integrations – Production-ready patterns – Active community and development
Execution Flow
1. Initialize agent with configuration
↓
2. Load environment and dependencies
↓
3. Prepare prompt with input data
↓
4. Invoke LLM through LangChain
↓
5. Parse and validate response
↓
6. Return formatted result
Key Decision Points: – Model selection (temperature, model type) – Prompt structure (few-shot vs zero-shot) – Error handling strategy – Output format validation
Performance Considerations
Computational Cost
- LLM Calls: Primary cost factor (based on tokens)
- Latency: Depends on model and input size
- Memory: Minimal for most patterns
Optimization Strategies
- Prompt Optimization: Minimize tokens while maintaining clarity
- Caching: Cache repeated queries/responses
- Model Selection: Use smaller models when appropriate
- Batch Processing: Group similar requests
- Streaming: For long responses, stream tokens
Cost Estimates (Approximate)
- GPT-3.5-turbo: ~$0.002 per request
- GPT-4: ~$0.03 per request
- GPT-4-turbo: ~$0.01 per request
Actual costs vary based on input/output token counts
Common Pitfalls to Avoid
1. Prompt Design Issues
❌ Vague or ambiguous instructions
✅ Clear, specific, testable prompts
2. Error Handling
❌ Ignoring API errors or rate limits
✅ Implement retry logic and graceful degradation
3. Token Management
❌ Exceeding model token limits
✅ Truncate inputs, chunk processing
4. Security
❌ Exposing API keys in code
✅ Use environment variables
5. Testing
❌ Only testing happy paths
✅ Test edge cases, failures, malformed inputs
6. Cost Management
❌ Unmonitored API usage
✅ Implement budgets, alerts, and monitoring
Further Improvements
Production Enhancements
- Monitoring: Add logging, metrics, tracing
- Validation: Input/output schema validation
- Rate Limiting: Implement request throttling
- Caching: Redis or in-memory cache for responses
- Async Support: Non-blocking operations for scalability
Feature Extensions
- Multi-Model Support: Switch between providers
- Fallback Models: Graceful degradation strategy
- Custom Tools: Domain-specific function calling
- Advanced Memory: Vector stores for long-term memory
- Evaluation: Automated testing and benchmarking
Code Quality
- Testing: Unit tests, integration tests
- Documentation: API docs, usage examples
- Type Safety: Stricter type hints
- Configuration: Externalized config files
Related Patterns
- Pattern 027
- Pattern 029
Pattern Families: – Reasoning Patterns (001-011) – Multi-Agent Patterns (012-019) – Tool Use Patterns (020-025) – Memory Patterns (026-031) – And many more…
Summary
The Working Memory pattern provides a powerful approach to building AI agents with specific capabilities. This implementation demonstrates:
✅ Key Strengths: – Clean, maintainable code structure – LangChain integration for composability – Type hints and documentation – Production-ready patterns
🎯 Best Use Cases: – Applications requiring this pattern’s characteristics – Systems needing structured agent behavior – Production environments requiring reliability
📚 Learning Resources: – LangChain Documentation: https://python.langchain.com/ – Pattern Documentation: See related patterns – OpenAI API Reference: https://platform.openai.com/docs/
File: 028_working_memory.py
Total Lines: 594
Pattern Category: AI Agent Design Pattern #028
Generated from actual implementation code
For educational and reference purposes