waken-groq 0.1.0

Groq Target adapter for Waken — routes waken Events to Groq's fast inference API.

waken-groq

A Groq Target adapter for Waken — "nginx for AI agents." Routes waken Events to Groq's chat completions API and returns the result as a waken Response.

Groq's API is shaped like OpenAI's Chat Completions API — its differentiator is inference speed (custom LPU hardware), not a different way of talking to it.

Install

pip install waken-groq

Usage

from waken import Runtime
from waken_groq import GroqAdapter

runtime = Runtime()
runtime.target("groq", GroqAdapter())
runtime.run()

GroqAdapter reads GROQ_API_KEY from the environment — that's the groq SDK's own convention (see AsyncGroq()), nothing waken-specific. Pass api_key=... (or any other AsyncGroq keyword) to override:

runtime.target("groq", GroqAdapter(model="llama-3.1-8b-instant", api_key="..."))

Sessions

Groq's chat completions API is stateless per call — there is no server-side session or thread, the caller resends the full conversation every time. So GroqAdapter maps a waken session_id to a plain list of chat messages that it grows and resends on every turn. This history lives only in the adapter's process memory — it is never persisted, and does not survive a process restart. An Event with no session_id sends a single one-off message and is never stored.

Development

git clone https://github.com/WakenHQ/waken-groq
cd waken-groq
pip install -e ".[dev]"
pytest

Tests mock groq.AsyncGroq entirely — no API key or network access needed to run the suite.

License

MIT