Stream events from orxhestra agents including token-by-token output, sub-agent events via AgentTool, and Runner streaming.
Scanned 5/27/2026
Install via CLI
openskills install NicolaiLassen/orxhestra---
name: agent-streaming
description: Stream events from orxhestra agents including token-by-token output, sub-agent events via AgentTool, and Runner streaming.
---
# Agent Streaming
All agents stream via `astream()`, yielding `Event` objects.
## Basic streaming
```python
from orxhestra.events.event import Event, EventType
async for event in agent.astream("Write about distributed systems"):
if event.type == EventType.AGENT_MESSAGE and event.partial:
print(event.text, end="", flush=True)
elif event.is_final_response():
print(f"\n[DONE] {event.text}")
```
## Sub-agent streaming via AgentTool
Sub-agent events stream through the parent in real-time. Events carry `branch` and `agent_name` fields.
```python
from orxhestra import LlmAgent
from orxhestra.tools.agent_tool import AgentTool
weather_agent = LlmAgent(name="WeatherAgent", model=model, tools=[get_weather])
travel_agent = LlmAgent(name="TravelAgent", model=model, tools=[get_attractions])
planner = LlmAgent(
name="TripPlanner",
model=model,
tools=[AgentTool(weather_agent), AgentTool(travel_agent)],
instructions="Use the sub-agents to plan a trip.",
)
async for event in planner.astream("Plan a trip to Copenhagen"):
if event.branch:
print(f" [{event.agent_name}] {event.text}", end="")
elif event.is_final_response():
print(f"\nFinal: {event.text}")
```
## How it works
1. `LlmAgent` creates an `asyncio.Queue` and sets `ctx.event_callback = queue.put_nowait`.
2. `AgentTool` calls `ctx.event_callback(event)` for each child event.
3. Events yield from the queue concurrently while tools run.
4. `event_callback` propagates through `ctx.derive()` for nested sub-agents.
Any custom tool can use `ctx.event_callback` to push events.
## With Runner
```python
async for event in runner.astream(
user_id="user-1",
session_id="session-1",
new_message="Write me a long essay.",
):
if event.is_final_response():
print(f"\n[DONE] {event.text}")
elif event.type == EventType.AGENT_MESSAGE and event.partial:
print(event.text, end="", flush=True)
```
No comments yet. Be the first to comment!
End-to-end marketing campaign planning and execution. Covers audience research, positioning, campaign angle definition, landing page copy, email sequences, social posts, ad copy, short-form video scripts, and content calendars. Use as the orchestration layer for multi-channel product launches.
Orchestrate multi-phase deep research with web search, memory retrieval, pattern matching, and synthesis into structured findings
Persistent memory systems for LLM conversations including short-term, long-term, and entity-based memory Use when: conversation memory, remember, memory persistence, long-term memory, chat history.
CodeTour `.tour`ファイルを作成 — ペルソナターゲット、ステップバイステップウォークスルー(実際のファイルとラインアンカー付き)。オンボーディングツアー、アーキテクチャウォークスルー、PRツアー、RCAツアー、構造化「これがどのように機能するかを説明」リクエストに使用。
showコマンド、コンフィグ階層、ワイルドカードマスク、ACL配置、インターフェースハイジーン、安全な変更ウィンドウ検証のためのCisco IOSおよびIOS-XEレビューパターン。