Signal System¶
LLMling-agent uses signals to enable observation and reaction to various events throughout the system. These signals are particularly useful for building UIs, monitoring agent behavior, and implementing custom event handling.
Registry Events¶
Several components provide a standard set of signals for monitoring changes to their collections. These signals are accessible through the events property:
# Available on any BaseRegistry
events.adding = Signal(object) # (key, ) - Before addition
events.added = Signal(object, object) # (key, value) - After addition
events.changing = Signal(object) # (key, ) - Before change
events.changed = Signal(object, object, object) # (key, old_value, value) - After change
events.removing = Signal(object) # (key, ) - Before removal
events.removed = Signal(object, object) # (key, value) - After removal
Components using BaseRegistry¶
AgentPool¶
Monitors agent additions and removals:
def on_agent_added(name: str, agent: Agent):
print(f"New agent {name} added to pool")
pool.events.added.connect(on_agent_added)
ToolManager¶
Tracks tool registration and modifications:
def on_tool_changed(name: str, old_info: ToolInfo, new_info: ToolInfo):
print(f"Tool {name} configuration changed")
agent.tools.events.changed.connect(on_tool_changed)
Agent Signals¶
The main Agent class provides signals for monitoring core agent activities:
message_received = Signal(ChatMessage[str]) # New message received
message_sent = Signal(ChatMessage) # Agent sent a message
tool_used = Signal(ToolCallInfo) # Tool was called
chunk_streamed = Signal(str, str) # Streaming chunk was emitted (chunk, message_id)
run_failed = Signal(str, Exception) # Run operation failed
agent_reset = Signal(AgentReset) # Agent state was reset
Provider Signals¶
Base provider signals:
tool_used = Signal(ToolCallInfo) # Tool execution
chunk_streamed = Signal(str, str) # Streaming response chunk
model_changed = Signal(object) # Model was changed
Conversation Manager Signals¶
Signals for conversation state changes:
Talk/Connection Signals¶
Signals for agent communication:
message_received = Signal(ChatMessage) # Original message
message_forwarded = Signal(ChatMessage) # After transformation
node_connected = Signal(object) # Message connection established
connection_added = Signal(Talk) # New connection created
Common Use Cases¶
Building UIs¶
class ConsoleUI:
def __init__(self, agent: Agent):
# Core agent activity
agent.message_sent.connect(self.display_message)
agent.tool_used.connect(self.display_tool_usage)
agent.chunk_streamed.connect(self.display_chunk)
# Tool management
agent.tools.events.added.connect(self.update_tool_display)
agent.tools.events.changed.connect(self.refresh_tool_status)
Monitoring and Logging¶
def setup_monitoring(pool: AgentPool):
# Monitor agent lifecycle
pool.events.added.connect(lambda name, agent: logger.info("Agent added: %s", name))
pool.events.removed.connect(lambda name, agent: logger.info("Agent removed: %s", name))
# Monitor tool usage across all agents
for agent in pool.values():
agent.tool_used.connect(lambda tool: logger.info("Tool used: %s", tool.name))