djeli

Python client for the djeli local LLM server.

djeli is a small, typed Python library for talking to a djeli serve process running on your machine. It lets you run GGUF language models locally and offline — no cloud, no API key, no data leaving your computer.

Installation

pip install djeli

Requires Python 3.8 or newer.

Prerequisite

This package is a client only. To use it, you need the djeli engine running on your machine.

1. Install the engine

Download the standalone binary for your OS (Linux, macOS, or Windows) from the official repository:

👉 https://github.com/bigdahaut/djeli

Follow the one-line install instructions in the repository's README.

2. Start the server

djeli serve

By default, it listens on http://127.0.0.1:11435.

Usage

from djeli import Djeli

client = Djeli()

answer = client.chat(
    model="qwen2.5-1.5b-instruct-q4_k_m",
    messages=[
        {"role": "system", "content": "You are a concise assistant."},
        {"role": "user", "content": "What is the capital of Kenya?"},
    ],
)
print(answer)
# The capital of Kenya is Nairobi.

Streaming

for chunk in client.chat(
    model="qwen2.5-1.5b-instruct-q4_k_m",
    messages=[{"role": "user", "content": "Write a haiku about Dakar."}],
    stream=True,
):
    print(chunk, end="", flush=True)

API Reference

Djeli(base_url="http://127.0.0.1:11435", timeout=300.0)

Create a client. timeout is the maximum time in seconds to wait for a response.

Argument Type Default Description

ase_url str "http://127.0.0.1:11435" URL of the local engine.

timeout float 300.0 Request timeout in seconds.

client.models() -> list[dict]

Return the models the engine can load. Each entry has id, object, created, and owned_by fields.

client.chat(model, messages, temperature=0.7, max_tokens=1024, stream=False)

Send a chat request.

Argument Type Default Description

model str — Model name: the .gguf filename without extension.

messages list[dict] — Conversation history. Each dict has role ("system", "user", "assistant") and content.

temperature float 0.7 Sampling temperature. 0.0 greedy, 0.7 balanced, 1.0+ creative.

max_tokens int 1024 Maximum generated tokens.

stream bool False Return an iterator of string chunks.

Returns str when stream=False, Iterator[str] when stream=True.

DjeliError

Raised on non-2xx responses, connection failures, or malformed responses. The message contains the engine's error text when available.

License

MIT