Loop Streaming Orchestrator¶

Repository: amplifier-module-loop-streaming

Token-level streaming orchestration that delivers LLM responses in real-time for interactive applications.

Contract¶

Field	Value
Module Type	`orchestrator`
Mount Point	`orchestrator`
Entry Point	`amplifier_module_loop_streaming:mount`

Behavior¶

Token-level streaming from provider
Real-time response delivery via hooks
Parallel tool execution — multiple tool calls execute concurrently
Deterministic context updates — results added in original call order
Interruptible generation
Extended thinking support for compatible providers

Configuration¶

[[orchestrators]]
module = "loop-streaming"
name = "streaming"
config = {
    max_iterations = -1,               # Maximum loop iterations (-1 = unlimited)
    extended_thinking = false,         # Enable extended thinking mode for supported models
    min_delay_between_calls_ms = 0,    # Minimum delay between provider calls in ms (0 = disabled)
    stream_delay = 0.0                 # Delay between streamed tokens in seconds (0.0 = no delay)
}

Config Reference¶

Key	Type	Default	Description
`max_iterations`	integer	`-1`	Maximum loop iterations (`-1` = unlimited)
`extended_thinking`	boolean	`false`	Enable extended thinking mode for supported models
`min_delay_between_calls_ms`	integer	`0`	Minimum delay between provider calls in milliseconds (`0` = disabled)
`stream_delay`	float	`0.0`	Delay between streamed tokens in seconds (artificial pacing; `0.0` = no delay)

Events Emitted¶

Event	Description
`execution:start`	Orchestrator execution began
`execution:end`	Orchestrator execution finished
`orchestrator:complete`	Loop completed (final event per `execute()` call)
`content_block:start`	Streaming block started
`content_block:delta`	Streaming token delta
`content_block:end`	Streaming block finished

Usage¶

In a Bundle¶

session:
  orchestrator:
    module: loop-streaming
    config:
      max_iterations: -1
      extended_thinking: false

Observing Streaming Tokens¶

from amplifier_core.events import CONTENT_BLOCK_DELTA

@coordinator.hooks.on(CONTENT_BLOCK_DELTA)
async def on_token(event_name, data):
    print(data.get("delta", ""), end="", flush=True)

Limiting Iterations¶

session:
  orchestrator:
    module: loop-streaming
    config:
      max_iterations: 5  # Stop after 5 agentic loop iterations

Extended Thinking¶

session:
  orchestrator:
    module: loop-streaming
    config:
      extended_thinking: true  # Requires provider with thinking support

Perfect For¶

Interactive CLI applications
Web UIs with progressive rendering
Long-form content generation
Agentic loops with parallel tool execution

Installation¶

amplifier module install loop-streaming

Or in a bundle:

session:
  orchestrator:
    module: loop-streaming
    source: git+https://github.com/microsoft/amplifier-module-loop-streaming@main