UI4AI 0.2.1

A Streamlit UI for LLM chat applications with persistence and chat history

UI4AI

A simple, lightweight, and plug-and-play Streamlit-based UI for LLM chatbot applications with ChatGPT-style features.

---

Features

• Plug in your own generate_response or streaming generate_response_stream function
• Built-in sidebar history and session management
• Welcome screen with optional suggestion chips when starting a new chat
• Custom avatars for user and assistant messages
• Optional: title generation, token counting, max history control
• Editable conversation titles and persistent sessions (survives refresh/restart)
• Optional conversation search

---

Installation

pip install UI4AI

---

Minimal usage (frontend only, no API key)

from UI4AI import run_chat

def generate_response(messages):
    return f"You said: {messages[-1]['content']}"

run_chat(generate_response=generate_response)

Run: streamlit run app.py

For OpenAI: use run_chat_openai() (requires API key).

---

Basic usage (customize with parameters)

Use run_chat() with your own generate_response:

from UI4AI import run_chat
from openai import OpenAI

client = OpenAI(api_key="<YOUR_API_KEY>")

def generate_response(messages):
    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=messages,
        temperature=0.7,
    )
    return response.choices[0].message.content or ""

run_chat(
    generate_response=generate_response,
    page_title="My Chatbot",
    header_title="My Chatbot",
)

Or use run_chat_openai() and override any parameter:

run_chat_openai(
    page_title="My Bot",
    system_prompt="You are a helpful assistant.",
    model="gpt-4o",
)

Run the app: streamlit run app.py

---

Examples

File	Description
examples/minimal_example.py	Frontend only, no API key
examples/simple_example.py	Echo bot with more responses
examples/openai_example.py	OpenAI (requires API key)
examples/base_example.py	Full template with all parameters documented

---

Streaming (typewriter effect)

Pass a generator that yields string chunks to get a ChatGPT-like streaming response:

from UI4AI import run_chat
from openai import OpenAI

client = OpenAI(api_key="<YOUR_API_KEY>")

def generate_response_stream(messages):
    stream = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=messages,
        stream=True,
    )
    for chunk in stream:
        if chunk.choices[0].delta.content:
            yield chunk.choices[0].delta.content

run_chat(
    generate_response_stream=generate_response_stream,
    page_title="Streaming Chat",
)

---

Welcome screen and avatars

run_chat(
    generate_response=my_response_fn,
    welcome_title="Welcome",
    welcome_message="How can I help you today?",
    suggestions=[
        "Explain quantum computing in simple terms",
        "Write a short poem",
        "Help me debug this code",
    ],
    user_avatar="🧑",
    assistant_avatar="🤖",
)

---

run_chat_openai parameters

Parameter	Type	Default	Description
api_key	str or None	None	OpenAI API key (or use OPENAI_API_KEY env / sidebar)
model	str	"gpt-4o-mini"	Model name
temperature	float	0.7	Sampling temperature
stream	bool	False	Use streaming responses
**kwargs	—	—	Any run_chat parameter (e.g. page_title, system_prompt)

---

run_chat parameter reference

Parameter	Type	Default	Description
generate_response	Callable[[List[Dict]], str] or None	None	Function that takes messages and returns the full response text.
generate_response_stream	Callable[[List[Dict]], Iterator[str]] or None	None	Generator that yields response chunks (used for streaming; takes priority over generate_response).
generate_title	Callable[[str], str] or None	None	Generates a conversation title from the first user message.
count_tokens	Callable[[List[Dict]], int] or None	None	Returns token count for the conversation (enables token display and max_history_tokens).
page_title	str	"AI Chat"	Browser tab title.
header_title	str	"UI4AI"	Sidebar header title.
byline_text	str	"Powered by Kethan Dosapati"	Byline under the header.
layout	str	"wide"	Streamlit layout: "wide" or "centered".
new_conversation_label	str	"➕ New Chat"	Label for the new conversation button.
chat_placeholder	str	"Ask me anything..."	Placeholder for the chat input.
spinner_text	str	"Thinking..."	Text shown while generating (non-streaming).
max_history_tokens	int or None	None	Max tokens to keep in context (requires count_tokens).
show_edit_options	bool	True	Show edit/delete in conversation menu.
primary_color	str	"#4f8bf9"	Primary UI color.
hover_color	str	"#f0f2f6"	Hover color.
date_grouping	bool	True	Group conversations by date in sidebar.
show_token_count	bool	True	Show token count per conversation.
max_title_length	int	25	Max length of conversation title in sidebar.
storage_path	str or None	None	Custom path for conversation JSON file.
system_prompt	str or None	None	System message added to each conversation.
enable_search	bool	False	Enable conversation search in sidebar.
user_avatar	str or None	None	Avatar for user (emoji, :material/icon_name:, or image URL).
assistant_avatar	str or None	None	Avatar for assistant.
welcome_title	str	"Welcome"	Title shown when there are no messages.
welcome_message	str	"How can I help you today?"	Message shown on welcome screen.
suggestions	List[str] or None	None	Optional suggestion chips on welcome screen.
setup_callback	Callable or None	None	Called after set_page_config (e.g. for API key prompt). Use when you need st.* before the main UI.

---

Additional features

• Title generation — Automatically generates a conversation title from the first message when generate_title is provided.
• Token counting — Displays total token count per conversation when count_tokens is provided.
• Max history — Use max_history_tokens with count_tokens to limit context length (older messages are truncated).
• Editable titles — Rename conversations from the sidebar menu.
• Persistent sessions — Conversations are saved to a JSON file and persist across refreshes and restarts.
• Sidebar history — Switch between past conversations in the sidebar.

---

License

MIT