ChatAssist

An advanced, provider-agnostic Python library for chatting with LLMs. Works with OpenAI, OpenRouter, a local Ollama server, or any endpoint that speaks the OpenAI-compatible /chat/completions format.

Made by تیروتیر — tirotir.ir.

What's new in 0.2.0

• Multi-provider support: provider="openai", "openrouter", "ollama", or a fully custom base_url
• Streaming responses: stream_message() yields text as it's generated
• Multi-turn context: pass history= to keep a real conversation going, not just single-shot messages
• System prompts: set a persona/instructions once, reused across turns
• API key optional: not required for local Ollama
• Env var support: CHATASSIST_API_KEY or OPENAI_API_KEY
• Fixed CLI packaging: chatassist-cli now actually installs as a command (this was broken in 0.1.0)
• Interactive CLI chat mode: just run chatassist-cli with no message for a REPL
• Markdown export for conversations, alongside JSON
• ChatGPTAPI is kept as a backward-compatible alias for existing code

Installation

pip install chatassist

Quick start

OpenAI

from chatassist import LLMClient

client = LLMClient(provider="openai", api_key="sk-...")
print(client.send_message("Hello!"))

OpenRouter

client = LLMClient(
    provider="openrouter",
    api_key="sk-or-...",
    model="anthropic/claude-3.5-sonnet",
)
print(client.send_message("Hello!"))

Local Ollama (no API key needed)

client = LLMClient(provider="ollama", model="llama3.2")
print(client.send_message("Hello!"))

Any custom OpenAI-compatible endpoint

client = LLMClient(
    base_url="https://my-llm-gateway.internal/v1/chat/completions",
    api_key="...",
    model="my-model",
)

Streaming

for chunk in client.stream_message("Write a short poem about the sea."):
    print(chunk, end="", flush=True)

Multi-turn conversations

from chatassist import LLMClient, Conversation

client = LLMClient(provider="ollama", model="llama3.2", system_prompt="You are terse.")
conversation = Conversation(system_prompt="You are terse.")

for user_msg in ["Hi!", "What's 2+2?", "And times 10?"]:
    reply = client.send_message(user_msg, history=conversation.as_api_messages())
    conversation.add_message("user", user_msg)
    conversation.add_message("assistant", reply)
    print(reply)

conversation.save_to_file("conversation.json")
conversation.export_markdown("conversation.md")

Command-line interface

# One-shot message
chatassist-cli "What is the capital of France?" --provider openai --api-key sk-...

# OpenRouter with a specific model
chatassist-cli "Summarize the plot of Dune" --provider openrouter --api-key sk-or-... --model anthropic/claude-3.5-sonnet

# Local Ollama, streamed
chatassist-cli "Tell me a joke" --provider ollama --model llama3.2 --stream

# Interactive REPL (omit the message)
chatassist-cli --provider ollama --model llama3.2

# Export the session
chatassist-cli "Hello" --export chat.json --export-markdown chat.md

Graphical User Interface

python -m chatassist.gui

On launch you'll get a small setup dialog to pick a provider, model, and API key (skip the key for Ollama) before the chat window opens.

20 examples, simple → advanced

The examples/ directory is a full lesson plan: 20 standalone, runnable scripts going from a one-line "hello world" up to streaming multi-turn chat, provider fallback, side-by-side model comparison, and a small FastAPI web service. See examples/README.md for the full index (in Persian, with English code/comments).

git clone https://github.com/tirotir-ir/chatassist.git
cd chatassist
pip install -e .
export OPENAI_API_KEY=sk-...
python examples/01_hello_world_openai.py

Development

pip install -e ".[dev]"
pytest tests/ -v   # runs fully offline, no API key needed

Configuration reference

Contributing

See CONTRIBUTING.md.

License

MIT License — see LICENSE.

ساخته و نگهداری‌شده توسط تیروتیر