Multi-Agent AI Systems - Orchestrating Collaborative Intelligence

1from typing import List, Dict, Any
2from dataclasses import dataclass
3from enum import Enum
4
5class AgentStatus(Enum):
6    IDLE = "idle"
7    WORKING = "working"
8    WAITING = "waiting"
9    COMPLETED = "completed"
10    FAILED = "failed"
11
12@dataclass
13class Agent:
14    id: str
15    role: str
16    capabilities: List[str]
17    status: AgentStatus = AgentStatus.IDLE
18    current_task: Optional[str] = None
19
20class Orchestrator:
21    def __init__(self):
22        self.agents: Dict[str, Agent] = {}
23        self.task_queue: List[Task] = []
24        self.completed_tasks: List[Task] = []
25
26    def register_agent(self, agent: Agent):
27        """Register an agent with the orchestrator"""
28        self.agents[agent.id] = agent
29        print(f"Agent {agent.id} registered with role: {agent.role}")
30
31    def assign_task(self, task: Task) -> Optional[Agent]:
32        """Assign task to most suitable available agent"""
33        # Find agents with required capabilities
34        capable_agents = [
35            agent for agent in self.agents.values()
36            if agent.status == AgentStatus.IDLE
37            and any(cap in agent.capabilities for cap in task.required_capabilities)
38        ]
39
40        if not capable_agents:
41            self.task_queue.append(task)
42            return None
43
44        # Select best agent (simple: first available)
45        selected_agent = capable_agents[0]
46        selected_agent.status = AgentStatus.WORKING
47        selected_agent.current_task = task.id
48
49        return selected_agent
50
51    def coordinate(self, task: Task):
52        """Coordinate multiple agents for complex task"""
53        # Break task into subtasks
54        subtasks = self.decompose_task(task)
55
56        # Assign subtasks to agents
57        assignments = {}
58        for subtask in subtasks:
59            agent = self.assign_task(subtask)
60            if agent:
61                assignments[subtask.id] = agent.id
62
63        # Monitor execution
64        results = self.monitor_execution(assignments)
65
66        # Aggregate results
67        return self.aggregate_results(results)
68
69    def decompose_task(self, task: Task) -> List[Task]:
70        """Break complex task into subtasks"""
71        # Use LLM to decompose task
72        decomposition_prompt = f"""
73        Task: {task.description}
74
75        Break this task into smaller, independent subtasks that can be
76        executed by different specialized agents.
77
78        For each subtask, specify:
79        - Description
80        - Required capabilities
81        - Dependencies on other subtasks
82
83        Return as structured JSON.
84        """
85
86        # Get decomposition from LLM
87        # Parse and create subtask objects
88        pass
89
90    def monitor_execution(self, assignments: Dict[str, str]) -> Dict[str, Any]:
91        """Monitor agent execution"""
92        results = {}
93        while assignments:
94            for task_id, agent_id in list(assignments.items()):
95                agent = self.agents[agent_id]
96
97                if agent.status == AgentStatus.COMPLETED:
98                    results[task_id] = agent.get_result()
99                    assignments.pop(task_id)
100                    agent.status = AgentStatus.IDLE
101
102                elif agent.status == AgentStatus.FAILED:
103                    # Handle failure - retry or reassign
104                    self.handle_failure(task_id, agent_id)
105                    assignments.pop(task_id)
106
107        return results
108

1import asyncio
2from typing import Set
3
4class Message:
5    def __init__(self, sender: str, receiver: str, content: Any, msg_type: str):
6        self.sender = sender
7        self.receiver = receiver
8        self.content = content
9        self.type = msg_type
10        self.timestamp = datetime.now()
11
12class P2PAgent:
13    def __init__(self, agent_id: str, role: str):
14        self.id = agent_id
15        self.role = role
16        self.peers: Set[str] = set()
17        self.inbox: asyncio.Queue = asyncio.Queue()
18        self.knowledge_base: Dict[str, Any] = {}
19
20    def connect_to_peer(self, peer_id: str):
21        """Establish connection with another agent"""
22        self.peers.add(peer_id)
23        self.broadcast_message({
24            "type": "peer_connected",
25            "peer_id": self.id
26        })
27
28    async def send_message(self, receiver: str, content: Any, msg_type: str):
29        """Send message to specific agent"""
30        message = Message(self.id, receiver, content, msg_type)
31        # Route message to receiver
32        await self.route_message(message)
33
34    async def broadcast_message(self, content: Any):
35        """Broadcast message to all peers"""
36        for peer_id in self.peers:
37            await self.send_message(peer_id, content, "broadcast")
38
39    async def process_messages(self):
40        """Process incoming messages"""
41        while True:
42            message = await self.inbox.get()
43            await self.handle_message(message)
44
45    async def handle_message(self, message: Message):
46        """Handle different message types"""
47        if message.type == "request":
48            response = await self.process_request(message.content)
49            await self.send_message(
50                message.sender,
51                response,
52                "response"
53            )
54
55        elif message.type == "share_knowledge":
56            self.update_knowledge(message.content)
57
58        elif message.type == "collaboration_request":
59            await self.handle_collaboration(message)
60
61    async def collaborate(self, task: str, required_roles: List[str]):
62        """Initiate collaboration with peers"""
63        # Find suitable peers
64        collaborators = await self.find_collaborators(required_roles)
65
66        # Propose collaboration
67        collaboration_id = f"collab_{uuid.uuid4()}"
68        proposals = []
69
70        for peer_id in collaborators:
71            proposal = await self.send_message(
72                peer_id,
73                {
74                    "collaboration_id": collaboration_id,
75                    "task": task,
76                    "initiator": self.id
77                },
78                "collaboration_request"
79            )
80            proposals.append(proposal)
81
82        # Wait for responses
83        responses = await self.collect_responses(collaboration_id)
84
85        # Form team
86        team = [peer for peer, accepted in responses.items() if accepted]
87
88        return CollaborationSession(collaboration_id, team, task)
89
90    async def find_collaborators(self, required_roles: List[str]) -> List[str]:
91        """Find peers with required capabilities"""
92        suitable_peers = []
93
94        for peer_id in self.peers:
95            # Query peer capabilities
96            response = await self.send_message(
97                peer_id,
98                {"query": "capabilities"},
99                "request"
100            )
101
102            if any(role in response.get("roles", []) for role in required_roles):
103                suitable_peers.append(peer_id)
104
105        return suitable_peers
106

1class HierarchicalAgent:
2    def __init__(self, agent_id: str, role: str, level: int):
3        self.id = agent_id
4        self.role = role
5        self.level = level  # 0 = worker, 1 = manager, 2 = director
6        self.subordinates: List[HierarchicalAgent] = []
7        self.manager: Optional[HierarchicalAgent] = None
8
9    def assign_subordinate(self, agent: 'HierarchicalAgent'):
10        """Add agent as subordinate"""
11        self.subordinates.append(agent)
12        agent.manager = self
13
14    def delegate_task(self, task: Task):
15        """Delegate task to subordinates"""
16        if not self.subordinates:
17            # Execute task directly
18            return self.execute(task)
19
20        # Decompose and assign to subordinates
21        subtasks = self.decompose_task(task)
22        results = []
23
24        for subtask, subordinate in zip(subtasks, self.subordinates):
25            result = subordinate.delegate_task(subtask)
26            results.append(result)
27
28        # Aggregate results
29        return self.aggregate_subordinate_results(results)
30
31    def report_to_manager(self, report: Dict[str, Any]):
32        """Report progress to manager"""
33        if self.manager:
34            self.manager.receive_report(self.id, report)
35
36    def receive_report(self, subordinate_id: str, report: Dict[str, Any]):
37        """Receive report from subordinate"""
38        # Process report
39        if report.get("needs_help"):
40            self.provide_assistance(subordinate_id, report["issue"])
41
42        # Aggregate and report upward if needed
43        if self.should_escalate(report):
44            self.report_to_manager({
45                "from": subordinate_id,
46                "issue": report,
47                "escalated_by": self.id
48            })
49

1from enum import Enum
2from dataclasses import dataclass
3
4class MessageType(Enum):
5    REQUEST = "request"
6    RESPONSE = "response"
7    INFORM = "inform"
8    QUERY = "query"
9    PROPOSE = "propose"
10    ACCEPT = "accept"
11    REJECT = "reject"
12
13@dataclass
14class AgentMessage:
15    sender: str
16    receiver: str
17    message_type: MessageType
18    content: Dict[str, Any]
19    conversation_id: str
20    timestamp: datetime
21
22class MessageBus:
23    def __init__(self):
24        self.queues: Dict[str, asyncio.Queue] = {}
25        self.message_history: List[AgentMessage] = []
26
27    def register_agent(self, agent_id: str):
28        """Register agent's message queue"""
29        self.queues[agent_id] = asyncio.Queue()
30
31    async def send(self, message: AgentMessage):
32        """Send message to recipient"""
33        if message.receiver not in self.queues:
34            raise ValueError(f"Agent {message.receiver} not registered")
35
36        await self.queues[message.receiver].put(message)
37        self.message_history.append(message)
38
39    async def receive(self, agent_id: str) -> AgentMessage:
40        """Receive message for agent"""
41        if agent_id not in self.queues:
42            raise ValueError(f"Agent {agent_id} not registered")
43
44        return await self.queues[agent_id].get()
45
46    def get_conversation(self, conversation_id: str) -> List[AgentMessage]:
47        """Retrieve full conversation"""
48        return [
49            msg for msg in self.message_history
50            if msg.conversation_id == conversation_id
51        ]
52

1from threading import Lock
2
3class Blackboard:
4    def __init__(self):
5        self.data: Dict[str, Any] = {}
6        self.locks: Dict[str, Lock] = {}
7        self.subscribers: Dict[str, List[callable]] = {}
8
9    def write(self, key: str, value: Any, agent_id: str):
10        """Write data to blackboard"""
11        if key not in self.locks:
12            self.locks[key] = Lock()
13
14        with self.locks[key]:
15            old_value = self.data.get(key)
16            self.data[key] = {
17                "value": value,
18                "written_by": agent_id,
19                "timestamp": datetime.now()
20            }
21
22            # Notify subscribers
23            self.notify_subscribers(key, value, old_value)
24
25    def read(self, key: str) -> Any:
26        """Read data from blackboard"""
27        return self.data.get(key, {}).get("value")
28
29    def subscribe(self, key: str, callback: callable):
30        """Subscribe to changes on a key"""
31        if key not in self.subscribers:
32            self.subscribers[key] = []
33        self.subscribers[key].append(callback)
34
35    def notify_subscribers(self, key: str, new_value: Any, old_value: Any):
36        """Notify subscribers of changes"""
37        if key in self.subscribers:
38            for callback in self.subscribers[key]:
39                callback(key, new_value, old_value)
40
41class BlackboardAgent:
42    def __init__(self, agent_id: str, blackboard: Blackboard):
43        self.id = agent_id
44        self.blackboard = blackboard
45
46    def contribute_knowledge(self, key: str, value: Any):
47        """Write knowledge to blackboard"""
48        self.blackboard.write(key, value, self.id)
49
50    def access_knowledge(self, key: str) -> Any:
51        """Access shared knowledge"""
52        return self.blackboard.read(key)
53
54    def watch_for_changes(self, key: str):
55        """Subscribe to knowledge changes"""
56        self.blackboard.subscribe(key, self.on_knowledge_updated)
57
58    def on_knowledge_updated(self, key: str, new_value: Any, old_value: Any):
59        """Handle knowledge updates"""
60        print(f"Agent {self.id}: {key} updated from {old_value} to {new_value}")
61        # React to changes
62

1from collections import Counter
2
3class VotingSystem:
4    def __init__(self, agents: List[Agent]):
5        self.agents = agents
6        self.votes: Dict[str, Any] = {}
7
8    async def call_vote(self, proposal: Dict[str, Any]) -> Dict[str, Any]:
9        """Conduct vote among agents"""
10        self.votes = {}
11
12        # Collect votes
13        vote_tasks = [
14            self.get_agent_vote(agent, proposal)
15            for agent in self.agents
16        ]
17        await asyncio.gather(*vote_tasks)
18
19        # Tally votes
20        return self.tally_votes()
21
22    async def get_agent_vote(self, agent: Agent, proposal: Dict[str, Any]):
23        """Get individual agent's vote"""
24        vote = await agent.vote_on_proposal(proposal)
25        self.votes[agent.id] = vote
26
27    def tally_votes(self) -> Dict[str, Any]:
28        """Count votes and determine outcome"""
29        vote_counts = Counter(self.votes.values())
30        total_votes = len(self.votes)
31
32        majority_threshold = total_votes // 2 + 1
33        winning_vote = vote_counts.most_common(1)[0]
34
35        return {
36            "outcome": winning_vote[0],
37            "vote_count": winning_vote[1],
38            "is_majority": winning_vote[1] >= majority_threshold,
39            "detailed_votes": dict(vote_counts)
40        }
41
42class VotingAgent:
43    async def vote_on_proposal(self, proposal: Dict[str, Any]) -> str:
44        """Agent evaluates proposal and votes"""
45        # Analyze proposal
46        analysis = await self.analyze_proposal(proposal)
47
48        # Make decision based on agent's goals and constraints
49        if analysis["benefits"] > analysis["costs"]:
50            return "approve"
51        elif analysis["benefits"] < analysis["costs"]:
52            return "reject"
53        else:
54            return "abstain"
55

1class ConsensusAgent:
2    def __init__(self, agent_id: str, initial_position: Any):
3        self.id = agent_id
4        self.position = initial_position
5        self.confidence = 1.0
6
7    async def negotiate(self, other_agents: List['ConsensusAgent'], rounds: int = 5):
8        """Negotiate with other agents to reach consensus"""
9        for round_num in range(rounds):
10            # Share position
11            positions = [agent.position for agent in other_agents] + [self.position]
12
13            # Evaluate positions
14            evaluation = await self.evaluate_positions(positions)
15
16            # Update position if convinced
17            if evaluation["should_update"]:
18                self.position = evaluation["new_position"]
19                self.confidence = evaluation["confidence"]
20
21            # Check for consensus
22            if self.check_consensus(other_agents):
23                return True
24
25        return False
26
27    async def evaluate_positions(self, positions: List[Any]) -> Dict[str, Any]:
28        """Evaluate other agents' positions"""
29        prompt = f"""
30        My current position: {self.position}
31        Other positions: {positions}
32
33        Should I update my position based on the arguments presented?
34        Consider:
35        1. Strength of arguments
36        2. Evidence provided
37        3. Consensus building
38
39        Return: should_update (bool), new_position (if updating), confidence (0-1)
40        """
41
42        # Get evaluation from LLM
43        # Return structured decision
44        pass
45
46    def check_consensus(self, other_agents: List['ConsensusAgent']) -> bool:
47        """Check if consensus is reached"""
48        positions = [agent.position for agent in other_agents] + [self.position]
49        # Check if all positions are similar enough
50        return len(set(positions)) == 1
51

1class ContractNetManager:
2    async def announce_task(self, task: Task, agents: List[Agent]):
3        """Announce task and collect bids"""
4        # Announce task
5        bids = await self.collect_bids(task, agents)
6
7        # Evaluate bids
8        winner = self.select_winner(bids)
9
10        # Award contract
11        if winner:
12            await self.award_contract(winner, task)
13            return winner
14
15        return None
16
17    async def collect_bids(self, task: Task, agents: List[Agent]) -> List[Bid]:
18        """Collect bids from capable agents"""
19        bids = []
20
21        for agent in agents:
22            if agent.can_perform(task):
23                bid = await agent.submit_bid(task)
24                if bid:
25                    bids.append(bid)
26
27        return bids
28
29    def select_winner(self, bids: List[Bid]) -> Optional[Agent]:
30        """Select best bid"""
31        if not bids:
32            return None
33
34        # Evaluate bids (cost, quality, time)
35        scored_bids = [
36            (bid, self.score_bid(bid))
37            for bid in bids
38        ]
39
40        winner = max(scored_bids, key=lambda x: x[1])
41        return winner[0].agent
42
43    def score_bid(self, bid: Bid) -> float:
44        """Score bid based on multiple criteria"""
45        # Normalize and weight different factors
46        cost_score = 1.0 / (bid.cost + 1)  # Lower cost is better
47        quality_score = bid.quality  # Higher quality is better
48        time_score = 1.0 / (bid.estimated_time + 1)  # Faster is better
49
50        return (
51            0.4 * cost_score +
52            0.4 * quality_score +
53            0.2 * time_score
54        )
55
56class BiddingAgent:
57    async def submit_bid(self, task: Task) -> Optional[Bid]:
58        """Evaluate task and submit bid"""
59        # Assess capability
60        capability = self.assess_capability(task)
61
62        if capability < 0.5:  # Not capable enough
63            return None
64
65        # Estimate cost and time
66        cost = self.estimate_cost(task)
67        time = self.estimate_time(task)
68
69        return Bid(
70            agent=self,
71            task=task,
72            cost=cost,
73            quality=capability,
74            estimated_time=time
75        )
76

1class CoalitionAgent:
2    def __init__(self, agent_id: str, capabilities: List[str]):
3        self.id = agent_id
4        self.capabilities = capabilities
5        self.current_coalition: Optional[Coalition] = None
6
7    async def form_coalition(self, task: Task, available_agents: List['CoalitionAgent']):
8        """Form coalition to accomplish task"""
9        required_capabilities = task.required_capabilities
10
11        # Find complementary agents
12        coalition_members = self.find_complementary_agents(
13            available_agents,
14            required_capabilities
15        )
16
17        # Propose coalition
18        coalition = Coalition(task, coalition_members)
19
20        # Get agreement from all members
21        if await coalition.get_consensus():
22            self.current_coalition = coalition
23            return coalition
24
25        return None
26
27    def find_complementary_agents(
28        self,
29        agents: List['CoalitionAgent'],
30        required_capabilities: List[str]
31    ) -> List['CoalitionAgent']:
32        """Find agents that together cover required capabilities"""
33        selected = [self]
34        covered_capabilities = set(self.capabilities)
35
36        for agent in agents:
37            if agent.id == self.id:
38                continue
39
40            # Check if agent adds new capabilities
41            new_capabilities = set(agent.capabilities) - covered_capabilities
42
43            if new_capabilities and len(selected) < 10:  # Max coalition size
44                selected.append(agent)
45                covered_capabilities.update(agent.capabilities)
46
47            # Check if all capabilities covered
48            if all(cap in covered_capabilities for cap in required_capabilities):
49                break
50
51        return selected if covered_capabilities.issuperset(required_capabilities) else []
52
53class Coalition:
54    def __init__(self, task: Task, members: List[CoalitionAgent]):
55        self.task = task
56        self.members = members
57        self.agreements: Dict[str, bool] = {}
58
59    async def get_consensus(self) -> bool:
60        """Get agreement from all coalition members"""
61        for member in self.members:
62            agreed = await member.agree_to_coalition(self)
63            self.agreements[member.id] = agreed
64
65        return all(self.agreements.values())
66
67    async def execute_task(self):
68        """Execute task collaboratively"""
69        # Assign subtasks to members
70        assignments = self.assign_subtasks()
71
72        # Execute in parallel
73        results = await asyncio.gather(*[
74            member.execute_subtask(subtask)
75            for member, subtask in assignments.items()
76        ])
77
78        # Aggregate results
79        return self.aggregate_results(results)
80

1class SelfOrganizingSystem:
2    def __init__(self, agents: List[Agent]):
3        self.agents = agents
4        self.environment = Environment()
5
6    def simulate(self, steps: int):
7        """Simulate system to observe emergent behavior"""
8        for step in range(steps):
9            # Each agent observes environment
10            for agent in self.agents:
11                observations = agent.observe(self.environment)
12
13                # Agent makes local decision
14                action = agent.decide(observations)
15
16                # Agent acts on environment
17                agent.act(action, self.environment)
18
19            # Environment updates
20            self.environment.update()
21
22            # Analyze emergent patterns
23            self.analyze_patterns(step)
24
25    def analyze_patterns(self, step: int):
26        """Detect emergent patterns in agent behavior"""
27        # Analyze agent positions, states, interactions
28        patterns = {
29            "clusters": self.detect_clusters(),
30            "coordination_level": self.measure_coordination(),
31            "efficiency": self.measure_efficiency()
32        }
33
34        print(f"Step {step}: {patterns}")
35
36    def detect_clusters(self) -> int:
37        """Detect agent clustering"""
38        # Implementation depends on agent properties
39        pass
40
41    def measure_coordination(self) -> float:
42        """Measure how well agents coordinate"""
43        # Analyze agent interactions and outcomes
44        pass
45

1# Scientific research collaboration
2research_system = MultiAgentSystem()
3research_system.add_agents([
4    ResearchAgent("literature_reviewer", "Literature Review"),
5    ResearchAgent("data_analyst", "Data Analysis"),
6    ResearchAgent("hypothesis_generator", "Hypothesis Generation"),
7    ResearchAgent("experiment_designer", "Experiment Design"),
8    ResearchAgent("writer", "Paper Writing")
9])
10
11result = await research_system.collaborate_on_task(
12    "Investigate the effects of XYZ on ABC"
13)
14

1# Supply chain optimization
2supply_chain = MultiAgentSystem()
3supply_chain.add_agents([
4    SupplierAgent("supplier_1"),
5    ManufacturerAgent("manufacturer_1"),
6    DistributorAgent("distributor_1"),
7    RetailerAgent("retailer_1"),
8    LogisticsAgent("logistics_1")
9])
10
11# Agents negotiate and coordinate automatically
12supply_chain.optimize_operations()
13

1# Traffic management system
2city_system = MultiAgentSystem()
3city_system.add_agents([
4    TrafficLightAgent(intersection_id) for intersection_id in intersections
5] + [
6    EmergencyVehicleAgent(vehicle_id) for vehicle_id in emergency_vehicles
7])
8
9# Agents coordinate to optimize traffic flow
10city_system.run_real_time_optimization()
11