ai-mirow-projects
/
swarms
mirror of https://github.com/kyegomez/swarms
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
Branches Tags
${ item.name }
Create tag ${ searchTerm }
Create branch ${ searchTerm }
from 'ba5c7d1630'
${ noResults }
swarms/docs/swarms/structs/election_swarm.md

881 lines
26 KiB
Raw Blame History

# ElectionSwarm
The `ElectionSwarm` is a sophisticated multi-agent orchestrator selection system that implements AGENTSNET communication protocols for democratic decision-making. It enables agent voters to elect the best orchestrator candidate through various coordination algorithms, providing a structured approach to leadership selection in multi-agent workflows.
## Overview
The ElectionSwarm follows a democratic election workflow pattern:
1. **Task Reception**: User provides an election task to the swarm
2. **Voter Registration**: Agent voters are registered with their preferences and expertise
3. **Candidate Presentation**: Orchestrator candidates present their platforms and capabilities
4. **Message Passing**: Voters communicate with neighbors using AGENTSNET protocols
5. **Voting Process**: Voters cast structured votes with detailed reasoning
6. **Result Determination**: Election algorithms determine the winning orchestrator
7. **Context Preservation**: All conversation history and voting records are maintained
## Architecture
```mermaid
graph TD
A[User Task] --> B[ElectionSwarm]
B --> C[Voter Registration]
B --> D[Candidate Registration]
C --> E[Agent Voters]
D --> F[Orchestrator Candidates]
E --> G[Message Passing Protocol]
F --> G
G --> H[AGENTSNET Communication]
H --> I[Voting Process]
I --> J[Election Algorithms]
J --> K[Consensus Building]
J --> L[Leader Election]
J --> M[Matching Algorithm]
J --> N[Coloring Algorithm]
J --> O[Vertex Cover]
K --> P[Election Result]
L --> P
M --> P
N --> P
O --> P
P --> Q[Selected Orchestrator]
```
## Key Features
| Feature | Description |
|---------|-------------|
| **AGENTSNET Protocols** | Implements distributed computing algorithms for coordination |
| **Message Passing** | Synchronous neighbor-to-neighbor communication in rounds |
| **Structured Voting** | Voters provide detailed reasoning, confidence, and candidate rankings |
| **Multiple Algorithms** | Consensus, Leader Election, Matching, Coloring, Vertex Cover |
| **Cost Tracking** | Budget management and token usage monitoring |
| **Caching System** | Response caching to avoid redundant LLM calls |
| **Lazy Loading** | Agents instantiated only when needed |
| **Parallel Execution** | Concurrent agent operations for efficiency |
| **Comprehensive Logging** | Detailed logging for debugging and monitoring |
| **Flexible Configuration** | Configurable election parameters and algorithms |
| **Voter Tool Calls** | Optional tool calls for detailed voting explanations |
## ElectionSwarm Constructor
| Parameter | Type | Default | Description |
|-----------|------|---------|-------------|
| `name` | `str` | `"ElectionSwarm"` | Name of the election swarm |
| `description` | `str` | `"Orchestrator selection with AGENTSNET algorithms"` | Description of the swarm purpose |
| `voters` | `List[VoterProfile]` | `None` | List of agent voters (defaults created if None) |
| `candidates` | `List[CandidateProfile]` | `None` | List of orchestrator candidates (defaults created if None) |
| `election_config` | `ElectionConfig` | `None` | Election configuration settings |
| `max_loops` | `int` | `1` | Maximum number of election loops |
| `output_type` | `OutputType` | `"dict-all-except-first"` | Output format type |
| `verbose` | `bool` | `False` | Enable verbose logging |
| `enable_lazy_loading` | `bool` | `True` | Enable lazy agent loading |
| `enable_caching` | `bool` | `True` | Enable response caching |
| `batch_size` | `int` | `25` | Batch size for parallel operations |
| `budget_limit` | `float` | `200.0` | Budget limit for cost tracking |
| `*args` | `Any` | - | Additional positional arguments |
| `**kwargs` | `Any` | - | Additional keyword arguments |
## Core Components
### VoterProfile
Represents an agent voter in the orchestrator selection system.
```python
@dataclass
class VoterProfile:
voter_id: str
name: str
voter_type: VoterType
preferences: Dict[str, Any] = field(default_factory=dict)
expertise_areas: List[str] = field(default_factory=list)
voting_weight: float = 1.0
agent: Optional[Agent] = None
is_loaded: bool = False
demographics: Dict[str, Any] = field(default_factory=dict)
past_voting_history: List[Dict[str, Any]] = field(default_factory=list)
neighbors: List[str] = field(default_factory=list)
coordination_style: str = "collaborative"
leadership_preferences: Dict[str, float] = field(default_factory=dict)
```
### VoterType
Enumeration of voter types in the election system.
```python
class VoterType(str, Enum):
INDIVIDUAL = "individual"
GROUP = "group"
EXPERT = "expert"
DELEGATE = "delegate"
```
### CandidateProfile
Represents an orchestrator candidate in the selection system.
```python
@dataclass
class CandidateProfile:
candidate_id: str
name: str
candidate_type: CandidateType
party_affiliation: Optional[str] = None
policy_positions: Dict[str, Any] = field(default_factory=dict)
campaign_promises: List[str] = field(default_factory=list)
experience: List[str] = field(default_factory=list)
agent: Optional[Agent] = None
is_loaded: bool = False
support_base: Dict[str, float] = field(default_factory=dict)
campaign_strategy: Dict[str, Any] = field(default_factory=dict)
leadership_style: str = "collaborative"
coordination_approach: Dict[str, Any] = field(default_factory=dict)
technical_expertise: List[str] = field(default_factory=list)
```
### CandidateType
Enumeration of candidate types in the election system.
```python
class CandidateType(str, Enum):
INDIVIDUAL = "individual"
COALITION = "coalition"
PARTY = "party"
MOVEMENT = "movement"
```
### ElectionAlgorithm
AGENTSNET algorithms for coordination.
```python
class ElectionAlgorithm(str, Enum):
CONSENSUS = "consensus"
LEADER_ELECTION = "leader_election"
MATCHING = "matching"
COLORING = "coloring"
VERTEX_COVER = "vertex_cover"
```
### VoterDecision
Structured output from voter agents.
```python
@dataclass
class VoterDecision:
voter_id: str
rationality: str
vote: VoteResult
confidence: float = 0.0
reasoning_factors: List[str] = field(default_factory=list)
candidate_rankings: Dict[str, int] = field(default_factory=dict)
tool_call_explanation: Optional[str] = None
timestamp: datetime = field(default_factory=datetime.now)
```
### VoteResult
Possible vote results.
```python
class VoteResult(str, Enum):
FOR = "for"
AGAINST = "against"
ABSTAIN = "abstain"
INVALID = "invalid"
```
### ElectionResult
Results of an election.
```python
@dataclass
class ElectionResult:
election_id: str
algorithm_used: ElectionAlgorithm
total_voters: int
total_candidates: int
votes_cast: int
winner: Optional[str] = None
vote_distribution: Dict[str, int] = field(default_factory=dict)
voter_decisions: List[VoterDecision] = field(default_factory=list)
consensus_reached: bool = False
rounds_to_consensus: int = 0
timestamp: datetime = field(default_factory=datetime.now)
```
### MessagePassingProtocol
AGENTSNET-inspired message-passing protocol.
```python
class MessagePassingProtocol:
def __init__(self, rounds: int = 5, synchronous: bool = True):
self.rounds = rounds
self.synchronous = synchronous
self.current_round = 0
self.message_history: List[Dict[str, Any]] = []
```
### CostTracker
Track costs and usage for budget management.
```python
@dataclass
class CostTracker:
total_tokens_used: int = 0
total_cost_estimate: float = 0.0
budget_limit: float = DEFAULT_BUDGET_LIMIT
token_cost_per_1m: float = 0.15
requests_made: int = 0
cache_hits: int = 0
```
## Voter Tool Calls
The ElectionSwarm supports optional tool calls that allow voter agents to provide detailed explanations of their voting decisions. When enabled, each voter agent will make a tool call to `explain_voting_decision` before casting their vote.
### Tool Call Features
- **Detailed Explanations**: Voters provide comprehensive reasoning for their choices
- **Structured Data**: Tool calls include voter name, ID, chosen candidate, and reasoning
- **Confidence Levels**: Voters specify their confidence in their decision
- **Alternative Considerations**: Voters explain why other candidates were not chosen
- **Key Factors**: Voters list the most important factors influencing their decision
### Tool Call Structure
```python
{
"voter_name": "Alice Johnson",
"voter_id": "voter_001",
"chosen_candidate": "John Progressive",
"voting_reasoning": "Detailed explanation of why this candidate was chosen",
"key_factors": ["healthcare policy", "environmental stance", "experience"],
"confidence_level": 0.85,
"alternative_considerations": "Considered Sarah Conservative but disagreed on healthcare"
}
```
### Enabling Tool Calls
```python
# Enable voter tool calls in configuration
config = ElectionConfig(
config_data={
"enable_voter_tool_calls": True,
# ... other config options
}
)
election = ElectionSwarm(
election_config=config,
# ... other parameters
)
```
## AGENTSNET Communication Protocols
### Message Passing Protocol
The ElectionSwarm implements AGENTSNET-inspired message-passing for distributed coordination:
- **Synchronous Communication**: Agents communicate in discrete rounds
- **Neighbor-Only Messaging**: Agents can only communicate with immediate neighbors
- **JSON Message Format**: Structured message exchange
- **Multi-Round Process**: Multiple rounds for information propagation
- **Final Decision**: Agents make final decisions after communication rounds
### Communication Flow
```mermaid
sequenceDiagram
participant V1 as Voter 1
participant V2 as Voter 2
participant V3 as Voter 3
participant ES as ElectionSwarm
Note over ES: Round 1: Initial Communication
V1->>V2: Share candidate preferences
V2->>V1: Share candidate preferences
V2->>V3: Share candidate preferences
V3->>V2: Share candidate preferences
Note over ES: Round 2: Information Exchange
V1->>V2: Discuss reasoning factors
V2->>V1: Discuss reasoning factors
V2->>V3: Discuss reasoning factors
V3->>V2: Discuss reasoning factors
Note over ES: Round 3: Final Consensus
V1->>ES: Final vote with reasoning
V2->>ES: Final vote with reasoning
V3->>ES: Final vote with reasoning
```
## Election Algorithms
The ElectionSwarm implements five AGENTSNET-inspired algorithms for orchestrator selection. Each algorithm is designed for different coordination scenarios and communication patterns.
### 1. Consensus Algorithm
All agents must agree on a single orchestrator candidate through multiple rounds of communication.
- **Purpose**: Achieve unanimous agreement on orchestrator selection
- **Process**: Multiple rounds until consensus threshold is reached
- **Threshold**: Configurable consensus threshold (default: 60%)
- **Fallback**: Majority vote if consensus not reached
- **Implementation**: `_conduct_consensus_election()` method
- **Use Case**: When unanimous agreement is required
### 2. Leader Election Algorithm
Select a single leader orchestrator through distributed voting.
- **Purpose**: Elect one orchestrator to lead the multi-agent workflow
- **Process**: Direct voting with winner-takes-all mechanism
- **Complexity**: O(D) rounds where D is network diameter
- **Implementation**: `_conduct_leader_election()` method
- **Use Case**: When clear leadership hierarchy is needed
### 3. Matching Algorithm
Pair voters with compatible orchestrator candidates using maximal matching.
- **Purpose**: Create optimal voter-candidate pairings
- **Process**: Maximal matching to avoid conflicts
- **Complexity**: O(log* n) rounds
- **Implementation**: `_conduct_matching_election()` method
- **Use Case**: When specialized orchestrator-voter relationships are important
### 4. Coloring Algorithm
Group voters and candidates into compatible categories using graph coloring.
- **Purpose**: Assign roles such that neighbors have different roles
- **Process**: (Δ+1)-coloring where Δ is maximum node degree
- **Complexity**: O(log* n) rounds
- **Implementation**: `_conduct_coloring_election()` method
- **Use Case**: Role assignment with conflict avoidance
### 5. Vertex Cover Algorithm
Select minimal set of coordinator agents to cover all voters.
- **Purpose**: Choose minimal set of orchestrators to cover all voters
- **Process**: Minimal vertex cover selection
- **Complexity**: O(log* n) rounds
- **Implementation**: `_conduct_vertex_cover_election()` method
- **Use Case**: When resource-efficient coordination is needed
### Algorithm Selection
Choose the appropriate algorithm based on your coordination needs:
```python
# For unanimous decisions
ElectionAlgorithm.CONSENSUS
# For clear leadership
ElectionAlgorithm.LEADER_ELECTION
# For optimal pairings
ElectionAlgorithm.MATCHING
# For role assignment
ElectionAlgorithm.COLORING
# For minimal coordination
ElectionAlgorithm.VERTEX_COVER
```
## Usage Examples
### Basic Election Setup
```python
from swarms.structs.election_swarm import (
ElectionSwarm,
VoterProfile,
CandidateProfile,
VoterType,
CandidateType,
ElectionAlgorithm
)
# Create agent voters
voters = [
VoterProfile(
voter_id="agent_001",
name="Data Analysis Agent",
voter_type=VoterType.EXPERT,
preferences={
"technical_leadership": 0.9,
"data_processing": 0.8,
"workflow_efficiency": 0.6
},
expertise_areas=["data_science", "machine_learning"],
neighbors=["Research Agent", "Backend Agent"]
),
# ... more voters
]
# Create orchestrator candidates
candidates = [
CandidateProfile(
candidate_id="orchestrator_001",
name="Technical Lead Orchestrator",
candidate_type=CandidateType.INDIVIDUAL,
policy_positions={
"technical_leadership": "Focus on code quality and architecture",
"team_coordination": "Agile methodologies with clear planning"
},
campaign_promises=[
"Implement comprehensive code review processes",
"Establish automated testing pipelines"
],
leadership_style="transformational",
technical_expertise=["software_architecture", "devops"]
),
# ... more candidates
]
# Create election swarm
election = ElectionSwarm(
name="Multi-Agent Orchestrator Election",
voters=voters,
candidates=candidates,
verbose=True
)
```
### Running Elections
```python
# Run consensus election
result = election.run(
task="Select the best orchestrator for our multi-agent workflow",
election_type=ElectionAlgorithm.CONSENSUS,
max_rounds=5
)
# Run leader election
result = election.run(
task="Elect a leader to coordinate our development team",
election_type=ElectionAlgorithm.LEADER_ELECTION
)
# Run matching election
result = election.run(
task="Match voters with compatible orchestrator candidates",
election_type=ElectionAlgorithm.MATCHING
)
# Run election session with comprehensive analysis
session_result = election.run_election_session(
election_type=ElectionAlgorithm.CONSENSUS,
max_rounds=5
)
# Access detailed results
election_result = session_result["election_result"]
analysis = session_result["analysis"]
cost_stats = session_result["cost_statistics"]
```
### Board CEO Election Example
```python
# Create board of directors as voters
board_members = [
VoterProfile(
voter_id="director_001",
name="Sarah Chen - Technology Director",
voter_type=VoterType.EXPERT,
preferences={
"innovation": 0.9,
"technical_excellence": 0.8,
"digital_transformation": 0.7
},
expertise_areas=["technology", "innovation"],
coordination_style="data_driven",
leadership_preferences={
"technical_leadership": 0.9,
"innovation": 0.8
}
),
# ... more board members
]
# Create CEO candidates with diverse approaches
ceo_candidates = [
CandidateProfile(
candidate_id="ceo_001",
name="Alexandra Martinez - Innovation CEO",
candidate_type=CandidateType.INDIVIDUAL,
policy_positions={
"innovation_leadership": "Aggressive digital transformation",
"growth_strategy": "Rapid expansion with high-risk investments"
},
leadership_style="transformational",
coordination_approach={"innovation": 0.9, "risk_taking": 0.8}
),
# ... more CEO candidates
]
# Create board CEO election
ceo_election = ElectionSwarm(
name="Board of Directors CEO Election",
voters=board_members,
candidates=ceo_candidates,
verbose=True
)
# Run the election
result = ceo_election.run(
task="Elect the best CEO to lead our company",
election_type=ElectionAlgorithm.CONSENSUS
)
# Get comprehensive statistics
stats = ceo_election.get_election_statistics()
print(f"Election statistics: {stats}")
# Get swarm status
status = ceo_election.get_swarm_status()
print(f"Swarm status: {status}")
```
## Advanced Configuration
### Election Configuration
The ElectionSwarm uses a comprehensive configuration system with Pydantic validation.
```python
from swarms.structs.election_swarm import ElectionConfig, ElectionConfigModel
# Create configuration with validation
config = ElectionConfig(
config_data={
"election_type": "orchestrator_selection",
"max_candidates": 5,
"max_voters": 50,
"enable_consensus": True,
"enable_leader_election": True,
"enable_matching": True,
"enable_coloring": True,
"enable_vertex_cover": True,
"enable_caching": True,
"enable_voter_tool_calls": True,
"batch_size": 10,
"max_workers": 5,
"budget_limit": 100.0,
"default_model": "gpt-4o-mini",
"verbose_logging": True
}
)
# Validate configuration
errors = config.validate_config()
if errors:
print(f"Configuration errors: {errors}")
# Save configuration to file
config.save_config("election_config.yaml")
election = ElectionSwarm(
name="Advanced Election",
election_config=config,
voters=voters,
candidates=candidates
)
```
### ElectionConfigModel
The configuration model with validation constraints:
```python
class ElectionConfigModel(BaseModel):
election_type: str = Field(default="orchestrator_selection")
max_candidates: int = Field(default=5, ge=1, le=20)
max_voters: int = Field(default=100, ge=1, le=1000)
enable_consensus: bool = Field(default=True)
enable_leader_election: bool = Field(default=True)
enable_matching: bool = Field(default=True)
enable_coloring: bool = Field(default=True)
enable_vertex_cover: bool = Field(default=True)
enable_caching: bool = Field(default=True)
enable_voter_tool_calls: bool = Field(default=True)
batch_size: int = Field(default=25, ge=1, le=100)
max_workers: int = Field(default=10, ge=1, le=50)
budget_limit: float = Field(default=200.0, ge=0.0)
default_model: str = Field(default="gpt-4o-mini")
verbose_logging: bool = Field(default=False)
```
### Cost Tracking and Budget Management
```python
# Monitor election costs
election = ElectionSwarm(
budget_limit=50.0,
enable_caching=True
)
# Run election with cost monitoring
result = election.run(task="Select orchestrator")
# Get cost statistics
stats = election.get_election_statistics()
print(f"Total cost: ${stats['cost_statistics']['total_cost']:.2f}")
print(f"Cache hit rate: {stats['cost_statistics']['cache_hit_rate']:.1%}")
```
## Core Methods
### Step Method
Execute a single step of the ElectionSwarm.
```python
# Execute a single election step
result = election.step(
task="Select the best orchestrator for our team",
election_type=ElectionAlgorithm.CONSENSUS,
max_rounds=5
)
```
### Run Method
Run the ElectionSwarm for the specified number of loops.
```python
# Run multiple election loops
result = election.run(
task="Elect a leader for our development team",
election_type=ElectionAlgorithm.LEADER_ELECTION,
max_rounds=3
)
```
### Conduct Election
Conduct an election using the specified algorithm.
```python
# Conduct election directly
result = election.conduct_election(
election_type=ElectionAlgorithm.CONSENSUS,
max_rounds=5
)
```
### Run Election Session
Run a complete election session with comprehensive analysis.
```python
# Run election session with detailed analysis
session_result = election.run_election_session(
election_type=ElectionAlgorithm.CONSENSUS,
max_rounds=5
)
# Access results
election_result = session_result["election_result"]
analysis = session_result["analysis"]
cost_statistics = session_result["cost_statistics"]
```
## Swarm Management Methods
### Adding and Removing Participants
```python
# Add new voter
new_voter = VoterProfile(
voter_id="agent_007",
name="New Agent",
voter_type=VoterType.EXPERT
)
election.add_voter(new_voter)
# Add new candidate
new_candidate = CandidateProfile(
candidate_id="orchestrator_004",
name="New Orchestrator",
candidate_type=CandidateType.INDIVIDUAL
)
election.add_candidate(new_candidate)
# Remove participants
election.remove_voter("agent_007")
election.remove_candidate("orchestrator_004")
```
### Swarm Status and Statistics
```python
# Get swarm information
status = election.get_swarm_status()
print(f"Total participants: {status['total_participants']}")
print(f"Voters: {status['voters']}")
print(f"Candidates: {status['candidates']}")
print(f"Budget remaining: ${status['budget_remaining']:.2f}")
# Get comprehensive statistics
stats = election.get_election_statistics()
print(f"Swarm name: {stats['swarm_info']['name']}")
print(f"Cache size: {stats['cache_stats']['cache_size']}")
# Get swarm size
swarm_size = election.get_swarm_size()
print(f"Total swarm size: {swarm_size}")
# Get specific voter or candidate
voter = election.get_voter("voter_001")
candidate = election.get_candidate("candidate_001")
```
### Reset and Cleanup
```python
# Reset election swarm to initial state
election.reset()
# This clears:
# - Conversation history
# - Cost tracking
# - Message protocol
# - Cache
```
## Best Practices
### 1. Voter Design
- **Diverse Expertise**: Include voters with different specializations
- **Balanced Preferences**: Ensure varied preference profiles
- **Neighbor Connections**: Create meaningful neighbor relationships
- **Realistic Demographics**: Use realistic agent characteristics
### 2. Candidate Design
- **Clear Platforms**: Define distinct policy positions
- **Compelling Promises**: Create realistic campaign promises
- **Relevant Experience**: Include pertinent background experience
- **Leadership Styles**: Vary leadership and coordination approaches
### 3. Election Configuration
- **Appropriate Algorithms**: Choose algorithms matching your use case
- **Budget Management**: Set realistic budget limits
- **Caching Strategy**: Enable caching for repeated elections
- **Logging Level**: Use verbose logging for debugging
### 4. Performance Optimization
- **Lazy Loading**: Enable lazy loading for large swarms
- **Batch Processing**: Use appropriate batch sizes
- **Parallel Execution**: Leverage concurrent operations
- **Cost Monitoring**: Track and optimize token usage
## Error Handling
The ElectionSwarm includes comprehensive error handling:
```python
try:
result = election.run(
task="Select orchestrator",
election_type=ElectionAlgorithm.CONSENSUS
)
except ValueError as e:
print(f"Configuration error: {e}")
except Exception as e:
print(f"Election failed: {e}")
```
Common error scenarios:
- **No voters or candidates**: Ensure at least one participant
- **Invalid algorithm**: Use supported ElectionAlgorithm values
- **Budget exceeded**: Monitor and adjust budget limits
- **Agent loading failures**: Check agent configurations
## Integration with Other Swarms
The ElectionSwarm can be integrated with other Swarms architectures:
```python
# Use with BoardOfDirectorsSwarm
from swarms.structs.board_of_directors_swarm import BoardOfDirectorsSwarm
# Create board election
board_election = ElectionSwarm(
name="Board CEO Election",
voters=board_members,
candidates=ceo_candidates
)
# Elect CEO
ceo_result = board_election.run(
task="Elect CEO for board governance",
election_type=ElectionAlgorithm.CONSENSUS
)
# Use elected CEO with BoardOfDirectorsSwarm
board = BoardOfDirectorsSwarm(
chairman=ceo_result.winner,
# ... other board configuration
)
```
## Utility Functions
### Create Default Configuration
Create a default election configuration file.
```python
from swarms.structs.election_swarm import create_default_election_config
# Create default configuration file
create_default_election_config("my_election_config.yaml")
```
### Default Constants
The ElectionSwarm uses several default constants:
```python
DEFAULT_BUDGET_LIMIT = 200.0
DEFAULT_CONSENSUS_THRESHOLD = 0.6
DEFAULT_MAX_WORKERS = 10
DEFAULT_MAX_ROUNDS = 5
DEFAULT_BATCH_SIZE = 25
```
## Conclusion
The ElectionSwarm provides a robust, scalable solution for orchestrator selection in multi-agent systems. By implementing AGENTSNET communication protocols and multiple election algorithms, it enables sophisticated democratic decision-making processes that can adapt to various coordination requirements and communication complexities.
Key advantages:
- **Distributed Coordination**: Implements proven distributed computing algorithms
- **Flexible Architecture**: Supports various election types and configurations
- **Production Ready**: Includes cost tracking, caching, and comprehensive error handling
- **Swarms Integration**: Seamlessly integrates with other Swarms architectures
- **Real-world Applications**: Suitable for corporate governance, team leadership, and workflow orchestration
The ElectionSwarm represents a significant advancement in multi-agent coordination, providing the tools necessary for sophisticated democratic decision-making in complex agent-based systems.
Reference in new issue View Git Blame Copy Permalink