Skip to content

Quick Start

A 5‑minute on‑ramp to AetherGraph: install, start the sidecar server, and run your first @graph_fn.

1. Install

pip install aethergraph
# or, from source
# pip install -e .

Python: 3.10+

2. Start the sidecar server (one‑liner)

AetherGraph ships a lightweight sidecar that wires up core services (logger, artifacts, memory, KV, channels, etc.)

# quickstart_server.py
from aethergraph import start_server

url = start_server()

Run it:

python quickstart_server.py

You should see a default HTTP URL http://127.0.0.1:8745 printed with UI (if available) and API urls.

3. Your first graph function

@graph_fn turns an ordinary async Python function into a runnable graph entrypoint. If you include a context parameter, you get access to built‑in services like context.channel() and context.memory().

# quickstart_graph_fn.py
from aethergraph import graph_fn, NodeContext
from aethergraph import start_server

# 1) Start the sidecar so services are available
start_server()

# 2) Define a small graph function
@graph_fn(name="hello.world", inputs=["name"], outputs=["greeting"])
async def hello_world(name: str, *, context: NodeContext):
    # Use the channel to send a message (console by default)
    await context.channel().send_text(f"👋 Hello, {name}! Running graph…")

    # Do any Python you want here — call tools, query memory, etc.
    greeting = f"Hello, {name}. Nice to meet you from AetherGraph."

    # Return outputs as a dict (keys must match `outputs=[...]`)
    return {"greeting": greeting}

# 3) Run it (async wrapper provided)
if __name__ == "__main__":
    import asyncio
    async def main():
        res = await hello_world(name="Researcher")
        print("Result:", res)
    asyncio.run(main())

Run it:

python quickstart_graph_fn.py

You should see a console message from the channel and printed output like:

Result: {"greeting": "Hello, Researcher. Nice to meet you from AetherGraph."}

4. What just happened?

  • Sidecar server booted in the background and installed default services (channels, artifacts, memory, KV, logger).
  • @graph_fn built a tiny task graph from your function and executed it.
  • context.channel() used the default channel (console) to emit a message.

Tip: You can override the channel at call‑site with context.channel(channel_key=...), once you’ve configured adapters like Slack, Telegram or Aethergraph UI.

5. Next steps

  • Add tools with @tool to wrap reusable steps and surface inputs/outputs.
  • Use @graphify for fan‑in / fan‑out graph construction when the body is mostly tool calls.
  • Explore artifacts (context.artifacts()), memory (context.memory()), and RAG (context.rag())
  • Config .env file to integrate external channel and llm features