open-taranis

Python framework for AI agents logic-only coding with streaming, tool calls, and multi-LLM provider support.

Only the "fairly stable" versions are published on PyPi, but to get the latest experimental versions, clone this repository and install it !

Installation

pip install open-taranis --upgrade

For package on PyPi

or

git clone https://github.com/SyntaxError4Life/open-taranis && cd open-taranis/ && pip install .

For last version

Quick Start

Simplest

import open_taranis as T

client = T.clients.openrouter() # API_KEY in env_var

messages = [
    T.create_user_prompt("Tell me about yourself")
]

stream = T.clients.openrouter_request(
    client=client,
    messages=messages,
    model="nvidia/nemotron-3-nano-30b-a3b:free", 
)

print("assistant : ",end="")
for token, tool, tool_bool in T.handle_streaming(stream) : 
    if token :
        print(token, end="")

To create a simple display using gradio as backend

import open_taranis as T
import open_taranis.web_front as W
import gradio as gr

gr.ChatInterface(
    fn=W.chat_fn_gradio(
    client=T.clients.openrouter(), # API_KEY in env_var
    request=T.clients.openrouter_request,
    model="nvidia/nemotron-3-nano-30b-a3b:free",
    _system_prompt="You are an agent named **Taranis**"
).create_fn(),
    title="web front"
).launch()

Make a simple agent with a context windows on the 6 last turns

import open_taranis as T

class Agent(T.agent_base):
    def __init__(self):
        super().__init__()

        self.client = T.clients.openrouter()
        self._system_prompt = [T.create_system_prompt(
            "You're an agent nammed **Taranis** !"
        )]


    def create_stream(self):
        return T.clients.openrouter_request(
            client=self.client,
            messages=self._system_prompt+self.messages,
            model="nvidia/nemotron-3-nano-30b-a3b:free"
        )

    def manage_messages(self):
        self.messages = self.messages[-12:] # Each turn have 1 user and 1 assistant

My_agent = Agent()

while True :
    prompt = input("user : ")

    print("\n\nagent : ", end="")

    for t in My_agent(prompt):
        print(t, end="", flush=True)
    
    print("\n\n","="*60,"\n")

---

Use the commands :

• taranis help : in the name...
• taranis update : upgrade the framework
• taranis open : open the TUI

The TUI :

• /help to start

Documentation :

• Base of the docs (coding some things before the real docs)

Roadmap

• v0.0.1: start
• v0.0.x: Add and confirm other API providers (in the cloud, not locally)
• v0.1.x: Functionality verifications in examples
• > v0.2.0: Add features for logic-only coding approach, start with agent_base
• v0.3.x: Add a full agent in TUI and upgrade web client deployments
• The rest will follow soon.

Changelog

v0.0.x : The start

• v0.0.4 : Add xai and groq provider
• v0.0.6 : Add huggingface provider and args for clients.veniceai_request

v0.1.x : Gradio, commands and TUI

• v0.1.0 : Start the docs, add update-checker and preparing for the continuation of the project...
• v0.1.1 : Code to deploy a frontend with gradio added (no complex logic at the moment, ex: tool_calls)
• v0.1.2 : Fixed a display bug in the web_front and experimentally added ollama as a backend
• v0.1.3 : Fixed the memory reset in the web_front and remove ollama module for openai front (work 100 times better)
• v0.1.4 : Fixed web_front for native use on huggingface, as well as handle_streaming which had tool retrieval issues
• v0.1.7 : Added a TUI and commands, detection of env variables (API keys) and tools in the framework

v0.2.x : Agents

• v0.2.0 : Adding agent_base

Advanced Examples

• tools call in a JSON database
• tools call in a HR JSON database in multi-rounds
• simple search agent with Brave API

Links

• PyPI
• GitHub Repository