v0.1.0 · self-hostable · MIT

Self-hostable LLM client
for restaurant tech.

A thin Python library that talks to any OpenAI-compatible /v1/chat/completions endpoint — with menu Q&A, dish classification, and order parsing built on top. No vendor lock-in. No egress you don't control.

quickstart.py
from chenki import ChenkiClient

client = ChenkiClient()            # free HF Space by default
answer = client.ask_about_menu(
    menu, "any nut-free desserts?"
)
print(answer.text)

Restaurant-domain helpers

Not another chat wrapper.

The primitives a restaurant actually ships — typed, testable, and built on top of any model you point it at.

🍽️
ask_about_menu

Menu Q&A

Answer guest questions against your menu — allergens, pairings, substitutions — grounded in your data, not the model's guesses.

🏷️
classify_dish

Dish classification

Tag dishes by cuisine, course, dietary flags. Returns typed dataclasses.

🧾
parse_order_text

Order parsing

Turn free-text orders into structured line items.

🛡️
prompt-injection-safe

Safe by default

Helpers wrap user text in guarded prompts, so a hostile menu note can't hijack the model.

Get started

Install in one line.

One runtime dependency (httpx). Python 3.10+.

$ pip install chenki

30-second example

from chenki import ChenkiClient, Message

client = ChenkiClient()
reply = client.chat([Message(role="user", content="Hello!")])
print(reply.text)

Live backend

Try it right here.

The default backend is a free Hugging Face Space running Qwen 2.5 1.5B Instruct (Q4_K_M) behind llama.cpp. Cold start ~30 s; ~20 tok/s after.

brianchenhao-chenki-llm.hf.space

Embedded from huggingface.co/spaces/brianchenhao/chenki-llm

Architecture

How it works.

01

Install the client

pip install chenki. One runtime dep (httpx). No vendor SDK lock-in.

02

Point at any OpenAI-compatible endpoint

ChenkiClient() defaults to the free chenki-llm Space. Or pass your own — Ollama, vLLM, llama.cpp, LM Studio, or OpenAI itself.

03

Use the restaurant helpers

ask_about_menu, classify_dish, parse_order_text — typed dataclasses, prompt-injection-safe by default.

Your app

Adopter's Python

  • ChenkiClient
  • ask_about_menu
  • classify_dish

chenki-llm · HF Space

Any OpenAI backend

  • Qwen 2.5 1.5B Q4_K_M
  • llama.cpp :server
  • OpenAI-compatible API

Self-hosters deploy their own Space with the included server/Dockerfile — same image, same API, zero third-party egress.

chenki.com · coming soon

Be first in line.

Hosted playground, managed endpoints, and a proper dashboard are on the way. Drop your email and we'll let you in early.

No spam. One email at launch, then nothing else.