streaming

Streaming Examples

This directory contains comprehensive examples demonstrating the streaming capabilities of the Agent SDK.

Overview

The Agent SDK supports real-time streaming from multiple LLM providers:

• Anthropic Claude: Full SSE support with thinking events
• OpenAI GPT: Delta streaming with structured outputs and o1 reasoning

Examples Included

1. Basic LLM Streaming

Demonstrates direct streaming from LLM providers with real-time token delivery.

2. Agent Streaming

Shows how agents can stream responses while maintaining context and memory.

3. Streaming with Tools

Illustrates tool execution during streaming, with progress events for tool calls.

4. Advanced Streaming Features

Showcases advanced features like custom buffer sizes, thinking events, and metrics.

Prerequisites

Install dependencies:

go mod tidy

Set up API keys for your chosen provider:

For Anthropic:

export ANTHROPIC_API_KEY=your_anthropic_key
export LLM_PROVIDER=anthropic

For OpenAI:

export OPENAI_API_KEY=your_openai_key
export LLM_PROVIDER=openai

Running the Examples

Run all examples:

go run main.go

Example Output

🚀 Agent SDK Streaming Examples
===============================

Using provider: anthropic

📡 Example 1: Basic LLM Streaming
--------------------------------
Starting LLM streaming...
Response: Quantum computing is a revolutionary approach to computation that harnesses the principles of quantum mechanics...
[Stream completed]

🤖 Example 2: Agent Streaming
-----------------------------
Starting agent streaming...
Agent Response: Machine learning works through several key steps...
🤔 [Thinking: I should break this down into clear, digestible steps...]
[Agent streaming completed]

🛠️  Example 3: Streaming with Tools
----------------------------------
Starting streaming with tools...
Response: I'll help you calculate compound interest step by step...
🔧 [Tool Call #1: calculator]
   Arguments: {"operation": "power", "a": 1.0125, "b": 40}
   Status: executing
✅ [Tool Result: 1.6436186844245104]
[Streaming with tools completed - 3 tools used]

⚡ Example 4: Advanced Streaming Features
----------------------------------------
Starting advanced streaming with custom configuration...
[Stream started at 14:23:45.123]
Response: Let me think through the implications of quantum entanglement...
🧠 [Thinking #1: Quantum entanglement is a phenomenon where particles become correlated...]
📊 [Usage info: {input_tokens: 45, output_tokens: 312, total_tokens: 357}]
[Stream completed - Duration: 3.2s, Events: 45, Content: 1247 chars, Thinking events: 3]

✅ All streaming examples completed!

Key Features Demonstrated

Event Types

• Content Events: Real-time text streaming
• Thinking Events: Reasoning process visibility (Anthropic)
• Tool Events: Tool call progress and results
• Error Events: Robust error handling
• Completion Events: Stream lifecycle management

Configuration Options

• Buffer Sizes: Configurable for performance tuning
• Thinking Inclusion: Toggle reasoning visibility
• Tool Progress: Control tool execution feedback
• Memory Integration: Context preservation across streams

Advanced Capabilities

• Multi-provider Support: Seamless switching between LLM providers
• Structured Outputs: JSON schema streaming (OpenAI)
• o1 Reasoning: Special handling for reasoning models (OpenAI)
• Context Management: Multi-tenancy and conversation tracking
• Metrics Collection: Performance monitoring and usage tracking

Architecture

User Application
       ↓
   Agent.RunStream()
       ↓
   LLM.GenerateStream()
       ↓
   SSE Parser (Provider-specific)
       ↓
   Channel Events
       ↓
   Event Processing
       ↓
   Real-time Output

Error Handling

The examples demonstrate robust error handling:

• Connection failures are gracefully handled
• Partial responses are preserved
• Stream interruptions trigger proper cleanup
• Context cancellation stops streams immediately

Performance Considerations

• Channel buffer sizes can be tuned for throughput
• Events include timestamps for latency analysis
• Memory usage is controlled through buffer management
• Context cancellation provides immediate cleanup

Next Steps

• Explore the gRPC streaming examples for service-to-service communication
• Check out the HTTP/SSE examples for browser integration
• Review the test files for additional streaming scenarios