ArielAI

Build chatbots in minutes using open-source models and Gradio.

ArielAI wraps HuggingFace Transformers, Ollama, and the HuggingFace Inference API into a single, clean interface — so you go from zero to a working chatbot in just a few lines of code.

---

Installation

pip install arielai

For local model inference with PyTorch:

pip install arielai[torch]

For Ollama support:

pip install arielai[ollama]

For HuggingFace Inference API (no GPU required):

pip install arielai[inference]

---

Quick Start

One-liner

import arielai
arielai.launch("microsoft/DialoGPT-medium")

Basic usage

from arielai import Chatbot

bot = Chatbot(
    model="microsoft/DialoGPT-medium",
    title="My Chatbot",
)
bot.launch()

Using a preset

from arielai import Chatbot

bot = Chatbot.from_preset("zephyr")
bot.launch()

---

Backends

ArielAI supports three backends:

Backend	backend=	Description
HuggingFace Transformers	"transformers"	Download and run models locally
Ollama	"ollama"	Local models via Ollama
HF Inference API	"inference"	Cloud-hosted, no GPU needed

Transformers (default)

bot = Chatbot(
    model="HuggingFaceH4/zephyr-7b-beta",
    backend="transformers",
    system_prompt="You are a helpful assistant.",
)
bot.launch()

Ollama

# First: install Ollama and pull a model
ollama pull llama3

bot = Chatbot(model="llama3", backend="ollama")
bot.launch()

HuggingFace Inference API

bot = Chatbot(
    model="HuggingFaceH4/zephyr-7b-beta",
    backend="inference",
    hf_token="hf_...",  # Optional for public models
)
bot.launch()

---

Presets

ArielAI ships with ready-to-use model presets:

Preset	Model	Backend
tiny	facebook/blenderbot-400M-distill	transformers
dialogue	microsoft/DialoGPT-medium	transformers
zephyr	HuggingFaceH4/zephyr-7b-beta	transformers
mistral	mistralai/Mistral-7B-Instruct-v0.2	transformers
ollama-llama3	llama3	ollama
ollama-mistral	mistral	ollama
ollama-gemma	gemma	ollama

from arielai import list_presets

for name, description in list_presets().items():
    print(f"{name}: {description}")

---

Full API Reference

Chatbot()

Parameter	Type	Default	Description
model	str	"microsoft/DialoGPT-medium"	HuggingFace model ID or Ollama model name
backend	str	"transformers"	"transformers", "ollama", or "inference"
title	str	"ArielAI Chatbot"	Chatbot window title
description	str	""	Description shown below the title
system_prompt	str	""	System instruction for the model
placeholder	str	"Ask me anything..."	Input box placeholder
max_new_tokens	int	512	Max tokens per response
temperature	float	0.7	Sampling temperature
examples	list[str]	None	Example messages shown in the UI
theme	str	"soft"	Gradio theme ("soft", "default", "monochrome", "glass")
streaming	bool	True	Stream responses token-by-token
device	str	None	Device for transformers ("cpu", "cuda", "mps")
load_in_8bit	bool	False	8-bit quantization (requires bitsandbytes)
load_in_4bit	bool	False	4-bit quantization (requires bitsandbytes)
hf_token	str	None	HuggingFace API token
ollama_host	str	"http://localhost:11434"	Ollama server URL

Chatbot.launch()

Parameter	Type	Default	Description
share	bool	False	Create a public Gradio share link
server_name	str	"0.0.0.0"	Bind address
server_port	int	None	Port (auto-selected if None)
inbrowser	bool	True	Open browser automatically

Chatbot.build()

Returns the gr.ChatInterface object without launching. Use this to embed ArielAI in an existing Gradio app or deploy to Hugging Face Spaces.

demo = bot.build()
demo.launch(...)

---

ChatUI — Polished Frontend

ChatUI wraps your chatbot in a branded, styled interface with a custom header, color themes, dark mode, and a footer. Still powered by Gradio, zero extra dependencies.

from arielai import Chatbot, ChatUI

bot = Chatbot(
    model="microsoft/DialoGPT-medium",
    streaming=True,
    examples=["Tell me something interesting.", "Write a short poem."],
)

ui = ChatUI(
    bot,
    brand_name="My AI Assistant",
    tagline="Powered by open-source AI",
    accent="indigo",        # color theme
    dark_mode=False,        # or True for dark
    footer_text="Built with ArielAI",
)

ui.launch()

ChatUI options

Parameter	Type	Default	Description
brand_name	str	"ArielAI"	Name shown in the header
tagline	str	"Powered by open-source AI"	Subtitle below the brand name
brand_logo	str	None	Path or URL to a logo image
accent	str	"indigo"	Color accent (see below)
dark_mode	bool	False	Dark color scheme
show_footer	bool	True	Show a footer below the chat
footer_text	str	"Built with ArielAI"	Footer label
footer_link	str	ArielAI GitHub	URL the footer text links to
extra_css	str	""	Extra raw CSS to inject

Available accents

indigo · emerald · rose · amber · blue · violet · cyan · slate

Or any hex color: accent="#ff6b6b"

from arielai import ChatUI
print(ChatUI.list_accents())

---

Advanced Usage

Embed in an existing Gradio app

import gradio as gr
from arielai import Chatbot

bot = Chatbot(model="microsoft/DialoGPT-medium")
chat_ui = bot.build()  # returns gr.ChatInterface

with gr.Blocks() as app:
    gr.Markdown("# My App")
    chat_ui.render()

app.launch()

Memory-efficient models (quantization)

bot = Chatbot(
    model="mistralai/Mistral-7B-Instruct-v0.2",
    load_in_4bit=True,  # Requires bitsandbytes + CUDA GPU
)
bot.launch()

Custom backend

from arielai import Chatbot
from arielai.backends import BaseBackend

class MyBackend(BaseBackend):
    def generate(self, message, history, **kwargs):
        return f"You said: {message}"

bot = Chatbot(backend_instance=MyBackend())
bot.launch()

---

Running Tests

pip install arielai[dev]
pytest

---

Contributing

Contributions are welcome! Please open an issue or pull request on GitHub.

---

License

MIT © ArielAI Contributors