Getting Started with Shannon: Deploy Multi-Agent Workflows in Minutes

Shannon promises production AI agents that actually work — but what does that look like in practice? I spun it up with Docker Compose and ran a real multi-agent workflow. Here's exactly how it went, from install to streaming events.

Step 1: One-Command Install

Shannon ships a single installer that pulls the Docker images, prompts for your LLM API keys, and starts everything:

curl -fsSL https://raw.githubusercontent.com/Kocoro-lab/Shannon/main/scripts/install.sh | bash

You need at least one provider key — OpenAI, Anthropic, or any OpenAI-compatible endpoint. Optional extras: SERPAPI_API_KEY for web search and FIRECRAWL_API_KEY for web fetch.

Step 2: Submit Your First Task

Once the gateway is listening on :8080, submitting a task is a single curl:

curl -X POST http://localhost:8080/api/v1/tasks \
  -H "Content-Type: application/json" \
  -d '{"query": "What is the capital of France?", "session_id": "demo"}'

What impressed me most is the observability — you can stream every orchestration event live:

curl -N "http://localhost:8080/api/v1/stream/sse?workflow_id=<task_id>"

Step 3: Python SDK & OpenAI Compatibility

The Python SDK feels clean and idiomatic:

from shannon import ShannonClient

with ShannonClient(base_url="http://localhost:8080") as client:
    handle = client.submit_task("What is the capital of France?", session_id="demo")
    result = client.wait(handle.task_id)
    print(result.result)

Even better for existing stacks: set OPENAI_API_BASE=http://localhost:8080/v1 and Shannon becomes a drop-in replacement — your current OpenAI code runs unchanged.

What Stands Out

• Token budget control — hard limits per task/agent with automatic model fallback; no more surprise bills
• Time-travel debugging — replay any workflow step-by-step instead of grepping logs
• Human approval workflows — pause agents for review before risky actions
• Multi-tenant isolation — WASI sandboxing and OPA policies for safe code execution

Verdict

Shannon is one of the few orchestration frameworks that feels built for production from day one. The Go/Rust core is fast, the observability is genuinely useful, and the OpenAI-compatible endpoint means zero migration cost. The desktop app is a nice bonus for watching agents in real time. If you're evaluating multi-agent infrastructure, Shannon deserves a serious look.