chatnificent 0.0.9

LLM chat app framework - Minimally complete. Maximally hackable

🗯️ Chatnificent

LLM chat app framework

Minimally complete. Maximally hackable.

Build production-ready, full-stack chat applications in minutes. Customize everything in hours.

Chatnificent is a Python framework built on Plotly's Dash designed to get your LLM chat applications up and running instantly, while providing a robust, decoupled architecture for unlimited customization.

Stop wrestling with UI components, state management, and backend integrations. Start building magnificent chat apps.

The Ethos

Frameworks should get out of your way.

• Minimally Complete: Out of the box, Chatnificent provides a fully functional, stateful, multi-user chat application with sensible defaults.
• Maximally Hackable: Every core pillar—the UI, the LLM provider, the database, the authentication, the RAG pipeline, and the core orchestration—is swappable. Customize or replace any part without fighting the framework.

Features

• LLM Agnostic: Built-in support for OpenAI, Anthropic, Gemini, Ollama, OpenRouter, DeepSeek, and any other LLM API.
• Flexible UI: Default Bootstrap layout, with built-in Mantine and Minimal (pure HTML) layouts. Easily customizable with any Dash components.
• Pluggable Storage: InMemory, File-system, and SQLite included. Easily extendable to Redis, Postgres, etc.
• Agentic Engine: The core engine manages multi-turn conversations and standardized tool calling across providers.
• Auth Ready: Abstracted authentication layer for easy integration. No-login anonymous user auth enabled by default.
• RTL Support: Automatic detection and rendering of Right-to-Left languages.
• Dash Native: Leverage the full power of Plotly's Dash to integrate complex data visualizations and analytics.

Installation

To get started quickly with the default UI (Bootstrap) and the default LLM provider (OpenAI):

pip install "chatnificent[default]"

export OPENAI_API_KEY="YOUR_API_KEY"

For a minimal installation (no UI libraries or LLM SDKs included):

pip install chatnificent

Quickstart: Hello World (3 Lines)

This is a complete, working chat application.

Create a file app.py:

from chatnificent import Chatnificent

app = Chatnificent()

if __name__ == "__main__":
    app.run(debug=True)

Run it:

python app.py

Open your browser to http://127.0.0.1:8050. That's it. You have a fully functional chat UI with conversation history, mobile responsiveness, and URL-based session management.

The Pillars of Hackability

Chatnificent's architecture is built around extensible Pillars. Every major function is handled by a dedicated component adhering to a strict interface.

Pillar	Description	Defaults	Included Implementations
LLM	The brain (API calls, parsing).	OpenAI (or Echo)	OpenAI, Anthropic, Gemini, OpenRouter, DeepSeek, Ollama, Echo
Layout	The look and feel (UI components).	Bootstrap (or Minimal)	Bootstrap, Mantine, Minimal (HTML)
Store	The memory (Persistence).	InMemory	InMemory, File, SQLite
Auth	The gatekeeper (User identification).	Anonymous	Anonymous, SingleUser
Engine	The orchestrator (Request lifecycle).	Synchronous	Synchronous
Tools	Tool/function calling capabilities.	NoTool	PythonTool, NoTool
Retrieval	RAG knowledge retrieval.	NoRetrieval	NoRetrieval
URL	URL parsing and routing.	PathBased	PathBased, QueryParams

You customize the app by injecting the implementations you need during initialization:

from chatnificent import Chatnificent
import chatnificent as chat

app = Chatnificent(
    llm=chat.llm.Anthropic(),
    store=chat.store.SQLite(db_path="conversations.db"),
    layout=chat.layout.Mantine()
)

Progressive Power: Swapping the Pillars

Let's evolve the "Hello World" example by swapping pillars.

Level 1: Swapping the LLM 🧠

Want to use Anthropic's Claude 3.5 Sonnet? Just swap the llm pillar.

(Requires pip install anthropic and setting ANTHROPIC_API_KEY)

from chatnificent import Chatnificent
import chatnificent as chat


app = Chatnificent(
    llm=chat.llm.Anthropic(model="claude-3-5-sonnet-20240620")
)

# Or try Gemini: app = Chatnificent(llm=chat.llm.Gemini())
# Or local Ollama: app = Chatnificent(llm=chat.llm.Ollama(model="llama3.1"))

Chatnificent handles the translation of message formats and tool-calling protocols automatically.

Level 2: Adding Persistent Storage

The default InMemory store is ephemeral. Let's use SQLite for persistence.

from chatnificent import Chatnificent
import chatnificent as chat

app = Chatnificent(
    store=store.SQLite(db_path="conversations.db")
)
# Or use the filesystem: store=chat.store.File(base_dir="./chat_data")

Conversations are now persisted across server restarts, and the sidebar automatically loads your history.

Level 3: Changing the Look and Feel 🎨

Don't want Bootstrap? Let's try the Mantine layout.

(Requires pip install dash-mantine-components)

from chatnificent import Chatnificent
import chatnificent as chat

app = Chatnificent(layout=chat.layout.Mantine())

# Or use the barebones HTML layout: layout=layout.Minimal()

Want a completely custom design? Implement the layout.Layout abstract base class. The framework ensures your custom layout integrates seamlessly, provided you include the required component IDs (e.g., input_textarea, messages_container, etc.).

Level 4: Custom Authentication

The default Anonymous auth isolates users by random user ID. You can easily implement custom logic.

from chatnificent import Chatnificent, auth

class HeaderAuth(auth.Auth):
    def get_current_user_id(self, **kwargs) -> str:
        from flask import request
        # Identify user based on a header (e.g., provided by an auth proxy)
        return request.headers.get("X-User-Id", "unknown_user")

app = Chatnificent(auth=HeaderAuth())

Level 5: The Engine (Advanced Orchestration)

The Engine orchestrates the entire request lifecycle: resolving the conversation, RAG retrieval, the agentic loop (Tools + LLM calls), and persistence.

The default Synchronous engine provides "hooks" (empty methods called at specific points) and "seams" (core logic methods) that you can override to deeply customize behavior without rewriting the core logic.

from chatnificent import Chatnificent
import chatnificent as chat
from typing import Any, Optional

# Create a custom engine by inheriting from the default
class CustomEngine(chat.engine.Synchronous):

    # 1. Override a HOOK to add monitoring/logging
    def _after_llm_call(self, llm_response: Any) -> None:
        # Example: Extract token usage if the LLM response object has a 'usage' attribute
        tokens = getattr(llm_response, 'usage', 'N/A')
        print(f"[MONITORING] LLM call complete. Tokens: {tokens}")

    # 2. Override a SEAM to modify core logic (e.g., prompt engineering)
    def _prepare_llm_payload(self, conversation, retrieval_context: Optional[str]):
        # Get the default payload (which already includes the context if present)
        payload = super()._prepare_llm_payload(conversation, retrieval_context)

        # Inject a custom system prompt if none exists
        if not any(m['role'] == 'system' for m in payload):
            payload.insert(0, {"role": "system", "content": "Be brief and professional."})
        return payload


# Initialize the app, passing the engine instance.
# Chatnificent's constructor will automatically bind the app reference to the engine.
app = Chatnificent(engine=CustomEngine())

Architecture Overview

How the pillars work together during a request:

1. User Input: The user submits a message via the Layout.
2. Callback Trigger: A Dash callback delegates the input to the Engine.
3. Context Resolution: The Engine uses Auth, URL, and Store to identify the user and load the conversation history.
4. Agentic Loop:
   • The Engine calls Retrieval to gather context (RAG).
   • The Engine sends the history and context to the LLM.
   • If the LLM requests a tool call, the Engine executes it via Tools and loops back.
   • If the LLM returns a final response, the loop exits.
5. Persistence: The Engine saves the updated conversation via the Store.
6. Rendering: The Engine formats the messages using the Layout and updates the client UI.

Building Your Own Pillars

The ultimate hackability comes from implementing your own pillars. Want to use MongoDB? Just implement the store.Store interface.

Example: Custom Storage Implementation

from chatnificent import Chatnificent
import chatnificent as chat
from typing import Optional, List

class MongoDBStore(chat.store.Store):
    def __init__(self, connection_string):
        # Initialize MongoDB client...
        print(f"Connecting to MongoDB at {connection_string}...")
        pass

    def load_conversation(self, user_id: str, convo_id: str) -> Optional[Conversation]:
        # Implement loading logic...
        return None

    # Implement the other required methods...
    def save_conversation(self, user_id: str, conversation: Conversation):
        pass
    def list_conversations(self, user_id: str) -> List[str]:
        return []
    def get_next_conversation_id(self, user_id: str) -> str:
        return "1"

# Use your custom implementation
# app = Chatnificent(store=MongoDBStore(connection_string="mongodb://..."))