Occupying-Mars
1 month ago
parent
570f8a2020
commit
776aafc1c5
@ -0,0 +1,117 @@
|
||||
# Notification Manager
|
||||
|
||||
The `NotificationManager` is responsible for managing selective notifications for vector DB updates between agents. It determines which agents should be notified of specific updates based on their expertise areas and notification preferences.
|
||||
|
||||
## Overview
|
||||
|
||||
The notification system uses relevance scoring and importance thresholds to determine which agents should receive updates. This prevents overwhelming agents with irrelevant notifications and ensures efficient information flow.
|
||||
|
||||
## Classes
|
||||
|
||||
### UpdateMetadata
|
||||
|
||||
Metadata for vector DB updates.
|
||||
|
||||
```python
|
||||
class UpdateMetadata(BaseModel):
|
||||
topic: str # Topic of the update
|
||||
importance: float # Importance score (0-1)
|
||||
timestamp: datetime # When update occurred
|
||||
embedding: Optional[List[float]] = None # Vector embedding
|
||||
affected_areas: List[str] = [] # Areas affected by update
|
||||
```
|
||||
|
||||
### AgentProfile
|
||||
|
||||
Profile containing agent notification preferences.
|
||||
|
||||
```python
|
||||
class AgentProfile(BaseModel):
|
||||
agent_id: str # Unique agent identifier
|
||||
expertise_areas: List[str] # Areas of expertise
|
||||
importance_threshold: float = 0.5 # Min importance threshold
|
||||
current_task_context: Optional[str] # Current task being worked on
|
||||
embedding: Optional[List[float]] # Vector embedding of expertise
|
||||
```
|
||||
|
||||
### NotificationManager
|
||||
|
||||
Main class for managing notifications.
|
||||
|
||||
```python
|
||||
class NotificationManager:
|
||||
def __init__(self):
|
||||
self.agent_profiles: Dict[str, AgentProfile] = {}
|
||||
```
|
||||
|
||||
## Key Methods
|
||||
|
||||
| Method | Description |
|
||||
|--------|-------------|
|
||||
| `register_agent(profile: AgentProfile)` | Register an agent's notification preferences |
|
||||
| `unregister_agent(agent_id: str)` | Remove an agent's notification preferences |
|
||||
| `calculate_relevance(update_metadata: UpdateMetadata, agent_profile: AgentProfile) -> float` | Calculate relevance score between update and agent |
|
||||
| `should_notify_agent(update_metadata: UpdateMetadata, agent_profile: AgentProfile) -> bool` | Determine if agent should be notified |
|
||||
| `get_agents_to_notify(update_metadata: UpdateMetadata) -> List[str]` | Get list of agents to notify |
|
||||
|
||||
## Usage Example
|
||||
|
||||
```python
|
||||
from swarms.structs.notification_manager import NotificationManager, UpdateMetadata, AgentProfile
|
||||
from datetime import datetime
|
||||
|
||||
# Create notification manager
|
||||
manager = NotificationManager()
|
||||
|
||||
# Register an agent
|
||||
profile = AgentProfile(
|
||||
agent_id="financial_agent",
|
||||
expertise_areas=["finance", "stocks", "trading"],
|
||||
importance_threshold=0.6
|
||||
)
|
||||
manager.register_agent(profile)
|
||||
|
||||
# Create an update
|
||||
update = UpdateMetadata(
|
||||
topic="stock_market",
|
||||
importance=0.8,
|
||||
timestamp=datetime.now(),
|
||||
affected_areas=["finance", "trading"]
|
||||
)
|
||||
|
||||
# Get agents to notify
|
||||
agents_to_notify = manager.get_agents_to_notify(update)
|
||||
print(f"Agents to notify: {agents_to_notify}")
|
||||
```
|
||||
|
||||
## Relevance Calculation
|
||||
|
||||
The relevance score between an update and an agent is calculated using:
|
||||
|
||||
1. Topic/expertise overlap score (70% weight)
|
||||
- Measures overlap between agent's expertise areas and update's affected areas
|
||||
|
||||
2. Embedding similarity score (30% weight)
|
||||
- Cosine similarity between update and agent embeddings if available
|
||||
|
||||
The final relevance score determines if an agent should be notified based on:
|
||||
- Relevance score > 0.5
|
||||
- Update importance > agent's importance threshold
|
||||
|
||||
## Best Practices
|
||||
|
||||
1. Set appropriate importance thresholds
|
||||
- Higher thresholds (e.g. 0.8) for critical updates only
|
||||
- Lower thresholds (e.g. 0.3) for broader awareness
|
||||
|
||||
2. Define focused expertise areas
|
||||
- Use specific areas rather than broad categories
|
||||
- Include related areas for better matching
|
||||
|
||||
3. Update task context
|
||||
- Keep current_task_context updated for better relevance
|
||||
- Clear context when task complete
|
||||
|
||||
4. Monitor notification patterns
|
||||
- Adjust thresholds if agents receive too many/few updates
|
||||
- Refine expertise areas based on relevance scores
|
||||
@ -0,0 +1,128 @@
|
||||
# Orchestrator
|
||||
|
||||
The `Orchestrator` manages communication and coordination between multiple agents, particularly for selective vector DB update notifications. It acts as a central hub for registering agents and routing notifications.
|
||||
|
||||
## Overview
|
||||
|
||||
The orchestrator maintains a list of agents and their notification preferences, using the `NotificationManager` to determine which agents should receive specific updates.
|
||||
|
||||
```mermaid
|
||||
graph TD
|
||||
A[Orchestrator] --> B[NotificationManager]
|
||||
A --> C[Agent List]
|
||||
B --> D[Calculate Relevance]
|
||||
B --> E[Determine Recipients]
|
||||
C --> F[Handle Updates]
|
||||
F --> G[Notify Agents]
|
||||
```
|
||||
|
||||
## Class Definition
|
||||
|
||||
```python
|
||||
class Orchestrator:
|
||||
def __init__(self):
|
||||
self.notification_manager = NotificationManager()
|
||||
self.agents: List[Agent] = []
|
||||
```
|
||||
|
||||
## Key Methods
|
||||
|
||||
| Method | Description |
|
||||
|--------|-------------|
|
||||
| `register_agent(agent: Agent)` | Register an agent with the orchestrator |
|
||||
| `handle_vector_db_update(update_metadata: UpdateMetadata)` | Process a vector DB update and notify relevant agents |
|
||||
| `update_agent_task_context(agent_name: str, task_context: str)` | Update an agent's current task context |
|
||||
|
||||
## Usage Example
|
||||
|
||||
```python
|
||||
from swarms import Agent, Orchestrator
|
||||
from swarms.structs.notification_manager import UpdateMetadata
|
||||
from datetime import datetime
|
||||
|
||||
# Create orchestrator
|
||||
orchestrator = Orchestrator()
|
||||
|
||||
# Create and register agents
|
||||
agent1 = Agent(
|
||||
agent_name="FinancialAgent",
|
||||
expertise_areas=["finance", "stocks", "trading"],
|
||||
importance_threshold=0.6
|
||||
)
|
||||
|
||||
agent2 = Agent(
|
||||
agent_name="TechAgent",
|
||||
expertise_areas=["technology", "software", "AI"],
|
||||
importance_threshold=0.4
|
||||
)
|
||||
|
||||
orchestrator.register_agent(agent1)
|
||||
orchestrator.register_agent(agent2)
|
||||
|
||||
# Example vector DB update
|
||||
update = UpdateMetadata(
|
||||
topic="stock_market",
|
||||
importance=0.8,
|
||||
timestamp=datetime.now(),
|
||||
affected_areas=["finance", "trading"]
|
||||
)
|
||||
|
||||
# Handle update - only FinancialAgent will be notified
|
||||
orchestrator.handle_vector_db_update(update)
|
||||
```
|
||||
|
||||
## Best Practices
|
||||
|
||||
1. Agent Registration
|
||||
- Register all agents at initialization
|
||||
- Update agent profiles when expertise changes
|
||||
- Remove inactive agents
|
||||
|
||||
2. Update Handling
|
||||
- Include relevant metadata in updates
|
||||
- Set appropriate importance levels
|
||||
- Consider update timing
|
||||
|
||||
3. Task Context
|
||||
- Keep task contexts current
|
||||
- Clear contexts when tasks complete
|
||||
- Use contexts to improve notification relevance
|
||||
|
||||
4. Monitoring
|
||||
- Track notification patterns
|
||||
- Adjust thresholds as needed
|
||||
- Monitor agent responses to notifications
|
||||
|
||||
## Integration with Vector Databases
|
||||
|
||||
The orchestrator can be integrated with various vector databases:
|
||||
|
||||
```python
|
||||
from swarms_memory import ChromaDB
|
||||
from swarms import Agent, Orchestrator
|
||||
|
||||
# Initialize vector DB
|
||||
vector_db = ChromaDB(
|
||||
metric="cosine",
|
||||
output_dir="agent_knowledge"
|
||||
)
|
||||
|
||||
# Create agents with vector DB
|
||||
agent = Agent(
|
||||
agent_name="ResearchAgent",
|
||||
expertise_areas=["research", "analysis"],
|
||||
long_term_memory=vector_db
|
||||
)
|
||||
|
||||
# Register with orchestrator
|
||||
orchestrator = Orchestrator()
|
||||
orchestrator.register_agent(agent)
|
||||
```
|
||||
|
||||
## Error Handling
|
||||
|
||||
The orchestrator includes error handling for:
|
||||
- Invalid agent registrations
|
||||
- Malformed updates
|
||||
- Failed notifications
|
||||
- Missing agent profiles
|
||||
Loading…
Reference in new issue