Skip to content

README.md

Latest commit

 

History

History
84 lines (61 loc) · 2.54 KB

README.md

File metadata and controls

84 lines (61 loc) · 2.54 KB

Streamkeeper Examples

This directory contains example applications demonstrating Streamkeeper (DurableStreams) functionality.

Requirements

  • Elixir 1.19+
  • Erlang/OTP 27+

Examples

1. Simple Demo (simple_demo.exs)

A command-line demonstration of the core DurableStreams API showing:

  • Stream creation
  • Appending data
  • Reading data
  • Long-polling for updates
  • JSON mode
  • Stream cleanup

Run:

cd /path/to/streamkeeper
elixir examples/simple_demo.exs

Expected output:

============================================================
DurableStreams Simple Demo
============================================================

--- 1. Creating a Stream ---
   Created stream: demo-stream
   Content-Type: text/plain
   ...

--- 2. Appending Data ---
   Appended: "Hello, World!"
   ...

2. LLM Token Streaming (llm_streaming.exs)

The flagship example demonstrating the primary use case from the Durable Streams announcement: resumable AI token streaming.

Features demonstrated:

  • Resumable streaming - Disconnect mid-AI-response and resume without losing tokens
  • Multi-client broadcast - Multiple tabs watch the same AI response in real-time
  • Replay capability - Re-watch the entire response from the beginning
  • DurableStreams.LiveView helper - Shows how to use the DSLive module for easy integration
  • Multi-provider support - Works with Claude (Anthropic), GPT (OpenAI), or Demo mode

Run:

cd /path/to/streamkeeper
iex examples/llm_streaming.exs

Then open http://localhost:4000 in your browser.

API Keys (Optional):

The example works without any API keys using Demo mode, which simulates streaming with pre-defined text. For real AI responses, set one or both:

export ANTHROPIC_API_KEY=your-anthropic-key-here
export OPENAI_API_KEY=your-openai-key-here

If both are set, one is randomly chosen for each generation.

Try this:

  1. Enter a prompt and click "Generate Response"
  2. Mid-stream, click "Disconnect", then "Resume" → no tokens lost!
  3. Open another browser tab and click "Replay from Start" to watch the full response
  4. Open multiple tabs before generating → all tabs see tokens stream in real-time

Dependencies

These examples use Phoenix Playground to create single-file Phoenix applications without needing a full project setup.

The examples automatically install their dependencies via Mix.install/1 - no manual setup required.