Streaming

Real-time response streaming from LLM providers.

Overview

Streaming allows you to receive LLM responses in real-time, token by token, enabling responsive user interfaces and progressive content display.

result = robot.run(state: state) do |event|
  case event.type
  when :text_delta
    print event.text
  when :tool_call
    puts "\nCalling tool: #{event.name}"
  when :complete
    puts "\nDone!"
  end
end

Components

Component	Description
Context	Streaming context and state
Events	Event types and handling

Quick Start

Basic Streaming

robot.run(state: state) do |event|
  print event.text if event.type == :text_delta
end

With Network

network.run(state: state) do |event|
  case event.type
  when :robot_start
    puts "Robot #{event.robot_name} starting..."
  when :text_delta
    print event.text
  when :robot_complete
    puts "\nRobot #{event.robot_name} complete"
  end
end

Event Types

Event	Description
:start	Streaming started
:text_delta	Text chunk received
:tool_call	Tool being called
:tool_result	Tool result received
:robot_start	Robot execution started
:robot_complete	Robot execution finished
:complete	All streaming finished
:error	Error occurred

Callback Patterns

Proc/Lambda

callback = ->(event) {
  print event.text if event.type == :text_delta
}

robot.run(state: state, streaming: callback)

Block

robot.run(state: state) do |event|
  print event.text if event.type == :text_delta
end

Object with call

class StreamHandler
  def call(event)
    case event.type
    when :text_delta
      broadcast(event.text)
    when :error
      log_error(event.error)
    end
  end
end

robot.run(state: state, streaming: StreamHandler.new)

See Also

• Streaming Guide
• Robot
• Network