Skip to content

Latest commit

 

History

History

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 

README.md

Agent Framework Orchestrations

Orchestration patterns for Microsoft Agent Framework. This package provides high-level builders for common multi-agent workflow patterns.

Installation

pip install agent-framework-orchestrations --pre

Orchestration Patterns

SequentialBuilder

Chain agents/executors in sequence, passing conversation context along:

from agent_framework.orchestrations import SequentialBuilder

workflow = SequentialBuilder(participants=[agent1, agent2, agent3]).build()

# Preserve agent1 and agent2 as visible progress, while the default builder output remains Workflow Output.
workflow = SequentialBuilder(
    participants=[agent1, agent2, agent3],
    intermediate_output_from=[agent1, agent2],
).build()

ConcurrentBuilder

Fan-out to multiple agents in parallel, then aggregate results:

from agent_framework.orchestrations import ConcurrentBuilder

workflow = ConcurrentBuilder(participants=[agent1, agent2, agent3]).build()

HandoffBuilder

Decentralized agent routing where agents decide handoff targets:

from agent_framework.orchestrations import HandoffBuilder

workflow = (
    HandoffBuilder()
    .participants([triage, billing, support])
    .with_start_agent(triage)
    .build()
)

GroupChatBuilder

Orchestrator-directed multi-agent conversations:

from agent_framework.orchestrations import GroupChatBuilder

workflow = GroupChatBuilder(
    participants=[agent1, agent2],
    selection_func=my_selector,
    intermediate_output_from=[agent1, agent2],
).build()

MagenticBuilder

Sophisticated multi-agent orchestration using the Magentic One pattern:

from agent_framework.orchestrations import MagenticBuilder

workflow = MagenticBuilder(
    participants=[researcher, writer, reviewer],
    manager_agent=manager_agent,
    intermediate_output_from=[researcher, writer, reviewer],
).build()

Output Selection

Orchestration builders expose Workflow Output selection using participant names. The core rule is that output_from is an allow-list for Workflow Output, not a routing rule for every other participant output. Unselected participant payloads are hidden unless intermediate_output_from explicitly selects them as Intermediate Output.

  • output_from designates participant emissions as Workflow Output (type='output' events).
  • intermediate_output_from designates participant emissions as Intermediate Output (type='intermediate' events).

If neither list is provided, each builder uses its documented default Workflow Output contract. Sequential emits the last participant; Concurrent, GroupChat, and Magentic emit their aggregator/orchestrator/manager output; Handoff emits participants.

Selection Workflow Output Intermediate Output Hidden payloads
Omit both selections Builder default Workflow Output contract None Builder-specific non-output participant payloads
output_from="all" Every output-capable participant None None
output_from=[writer] Only writer None All other participant payloads
output_from=[writer], intermediate_output_from="all_other" Only writer Every output-capable participant not selected by output_from None
intermediate_output_from="all_other" None, except builder-internal default output executors where applicable Every output-capable participant Builder-internal plumbing payloads
output_from=[], intermediate_output_from="all_other" None, except builder-internal default output executors where applicable Every output-capable participant Builder-internal plumbing payloads
output_from=[writer], intermediate_output_from=[researcher, reviewer] Only writer researcher and reviewer Any other participant payloads

Invalid selections fail at construction or build time:

Invalid selection Why it fails
output_from="all_other" "all_other" is only valid for intermediate_output_from
intermediate_output_from="all" "all" is only valid for output_from
The same participant in both selections One payload cannot be both Workflow Output and Intermediate Output
Duplicate participant selections Duplicates are treated as configuration errors
Unknown participant selections Typos and missing participants are rejected
output_from=[], intermediate_output_from=[] Both explicit selections are empty

When an orchestration is wrapped with workflow.as_agent(), Workflow Output becomes normal response text. Intermediate Output becomes text_reasoning content so callers can inspect progress without changing .text behavior.

Documentation

For more information, see the Agent Framework documentation.