Agents
Multi-agent orchestration with sequential, parallel, hierarchical, and workflow patterns.
Quick Start
from praisonaiagents import Agent, AgentTeam
researcher = Agent(name="Researcher", instructions="Research topics")
writer = Agent(name="Writer", instructions="Write content")
agents = AgentTeam(agents=[researcher, writer], memory=True)
result = agents.start()
Parameters Table
Core Parameters
| Parameter | Type | Default | Description |
|---|
agents | List[Agent] | required | List of agents to orchestrate |
tasks | List[Task] | None | Tasks to execute (auto-generated if None) |
process | str | "sequential" | Execution pattern (see below) |
name | str | None | Name for this agent collection |
variables | Dict | None | Global variables for substitution |
manager_llm | str | None | LLM for manager agent (hierarchical) |
Process Types
| Process | Description |
|---|
"sequential" | Execute tasks in order |
"parallel" | Execute independent tasks concurrently |
"hierarchical" | Manager agent delegates to workers |
"workflow" | Follow task dependencies |
Consolidated Feature Params
| Parameter | Type | Default | Description |
|---|
memory | bool | MultiAgentMemoryConfig | False | Shared memory |
planning | bool | MultiAgentPlanningConfig | False | Planning mode |
context | bool | ContextConfig | False | Context management |
output | str | MultiAgentOutputConfig | None | Output config |
execution | str | MultiAgentExecutionConfig | None | Execution config |
hooks | MultiAgentHooksConfig | None | Event hooks |
autonomy | bool | AutonomyConfig | None | Autonomy settings |
knowledge | bool | List[str] | KnowledgeConfig | None | Knowledge/RAG |
guardrails | bool | Callable | GuardrailConfig | None | Validation |
web | bool | WebConfig | None | Web search/fetch |
reflection | bool | ReflectionConfig | None | Self-reflection |
Precedence Ladder
Resolution Order: Instance > Config > Array > Dict > String > Bool > DefaultSame precedence as Agent. Consolidated params propagate to underlying agents.
| Form | Example | When to Use |
|---|
| Bool | memory=True | Enable with defaults |
| String preset | output="verbose" | Use predefined config |
| Dict | memory={"backend": "redis"} | Custom config |
| Array + overrides | output=["verbose", {"stream": False}] | Preset + customization |
| Config instance | memory=MultiAgentMemoryConfig(...) | Full control |
Presets & Options
Output Presets (Multi-Agent)
| Preset | Description |
|---|
"silent" | Zero output (default) |
"status" | Tool calls + response, no timestamps |
"trace" | Full trace with timestamps |
"debug" | trace + metrics (no boxes) |
"verbose" | Rich panels with Markdown |
Execution Presets (Multi-Agent)
| Preset | max_iter | max_retries |
|---|
"fast" | 5 | 2 |
"balanced" | 10 | 5 |
"thorough" | 20 | 5 |
"unlimited" | 100 | 10 |
Methods
| Method | Description |
|---|
start() | Start synchronous execution |
astart() | Async execution |
run_all_tasks() | Execute all tasks |
add_task(task) | Add task dynamically |
get_task_status(task_id) | Get task status |
set_state(key, value) | Set shared state |
get_state(key, default) | Get shared state |
Common Recipes
Sequential (Default)
Tasks execute one after another in order:
agents = AgentTeam(
agents=[agent1, agent2],
tasks=[task1, task2, task3],
process="sequential"
)
Parallel
Independent tasks execute concurrently:
agents = AgentTeam(
agents=[agent1, agent2, agent3],
tasks=[independent_task1, independent_task2, independent_task3],
process="parallel"
)
Hierarchical
Manager agent delegates to worker agents:
agents = AgentTeam(
agents=[worker1, worker2, worker3],
tasks=[complex_task],
process="hierarchical",
manager_llm="gpt-4o-mini"
)
Workflow
Tasks follow explicit dependencies via context:
task1 = Task(description="Step 1", agent=agent1)
task2 = Task(description="Step 2", agent=agent2, context=[task1])
task3 = Task(description="Step 3", agent=agent3, context=[task1, task2])
agents = AgentTeam(
agents=[agent1, agent2, agent3],
tasks=[task1, task2, task3],
process="workflow"
)
Async Support
Full async/await support for non-blocking execution:
import asyncio
from praisonaiagents import Agent, Task, AgentTeam
async def main():
agents = AgentTeam(
agents=[agent1, agent2],
tasks=[task1, task2],
process="workflow"
)
result = await agents.astart()
print(result["task_results"])
asyncio.run(main())
Shared State
Agents can share state during execution:
agents = AgentTeam(agents=[agent1, agent2], tasks=[task1, task2])
# Set shared state
agents.set_state("api_key", "xxx")
agents.set_state("config", {"mode": "production"})
# Access in tasks via context
result = agents.start()
# Read state after execution
final_data = agents.get_state("collected_data")
Advanced Examples
With Memory
agents = AgentTeam(
agents=[agent1, agent2],
tasks=[task1, task2],
memory=True, # Enable shared memory
user_id="user123"
)
With Custom Completion Checker
def my_checker(task, output):
"""Custom validation for task completion."""
if "error" in output.lower():
return False, "Output contains errors"
return True, output
agents = AgentTeam(
agents=[agent1],
tasks=[task1],
completion_checker=my_checker
)
With Verbose Logging
agents = AgentTeam(
agents=[agent1, agent2],
tasks=[task1, task2],
verbose=2 # Debug level
)
See Also
