Tried apache/iggy Today: A Fast Message Backbone for AI Pipelines

Apache Iggy is a persistent message-streaming platform built for high throughput and low latency. Written in Rust, it gives developers a lightweight alternative for moving events between services, workers, agents, and real-time applications without immediately reaching for a heavier distributed stack.

The project gained +15 GitHub stars today, and the reason is easy to understand: AI systems increasingly need reliable event streams for token output, tool calls, telemetry, document processing, and background jobs. Iggy fits that architecture well. Producers can publish work quickly, while independent consumers process, retry, or scale those events without tightly coupling services.

A practical AI layout looks like this:

API gateway
    |
    v
Iggy stream: ai-requests
    |
    +--> prompt worker
    +--> tool-calling worker
    +--> audit and metrics consumer

For a quick OpenAI-compatible gateway test, route a worker through B-Lost’s Universal Relay:

curl https://b-lost.com/v1/chat/completions \
  -H "Authorization: Bearer $BLOST_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-fable-5",
    "messages": [
      {"role": "user", "content": "Summarize this event payload."}
    ],
    "stream": true
  }'

The key design choice is separation: Iggy handles durable message delivery, while the relay handles model access. That makes it easier to swap providers, add retries, and keep API credentials out of streaming clients. The same base URL can be configured in Cursor, Cline, Roo Code, Windsurf, Aider, LibreChat, or other OpenAI-compatible clients.

B-Lost’s native Anthropic /v1/messages support is also useful when prompts repeat across many streamed jobs. Prompt caching can reduce cache-hit costs by up to 90%, while the listed 20% discount helps keep experimentation affordable.

Iggy is not trying to be an AI framework. That is its strength: it provides a fast, focused transport layer for building one.