Version 2.0 Now Live

Intent Classification
That Does Things.

Ultron isn't just another chatbot. It's a sophisticated execution engine that understands your intent and bridges the gap between natural language and system actions.

Explore Capabilities View Sandbox
Python 3.10+
Neural Nets
Shell Exec
Secure IO

One-Line Install

Ready to automate your world? Deployment is as simple as a single command. Our script handles everything: virtual environments, dependencies, and core configuration.

L
M
W
Cross-platform compatibility
Stable Release
$ curl -sSL https://raw.githubusercontent.com/wansatya/ultron/main/install.sh | bash

Requires Python 3.9+ and an internet connection.

Core Capabilities

Multi-Platform

Control your system from Discord, Telegram, Slack, or WhatsApp simultaneously. One brain, multiple access points for seamless orchestration.

Dynamic Skills

Add new capabilities instantly. Drop a Python file in the skills directory to hot-load new features without restarting the core engine.

Intelligent Extraction

Highly accurate entity extraction pulled directly from natural language. Ultron identifies URLs, paths, and commands with surgical precision.

Simple but Powerful

1

Intent Mapping

User message is converted to an embedding and compared against predefined intent manifolds.

2

Entity Extraction

The engine automatically extracts parameters like file paths, URLs, or search queries from your conversational input.

3

Multi-Channel Delivery

Responses and logs are delivered back to you across all active sessions, from Telegram to Discord, in real-time.

ultron-engine --active
> show me the config file
[INFO] Intent Classified: file_read (Confidence: 0.98)
[INFO] Executing tool: read(file='config.json') 
{
  "intents": [...],
  "models": {
    "classifier": "distilbert"
  }
}

Frequently Asked Questions

What exactly is CampShure's Ultron?
Ultron is a specialized agentic framework focused on high-precision intent classification. Unlike general-purpose LLMs that might hallucinate or struggle with specific system tasks, Ultron is designed to map natural language directly to actionable tools with extreme reliability.
How does the intent classification work?
We use a DistilBERT-based model to generate sentence embeddings for every user query. These embeddings are then compared using cosine similarity against a database of intention centroids. This allows for near-instant classification and high confidence scores even with subtle phrasing variations.
Which messaging platforms are supported?
Ultron currently features primary integration with Telegram. Our roadmap includes upcoming support for WhatsApp, Discord, and Slack, allowing you to control your system from your favorite messaging apps.
Does Ultron require a high-end GPU?
Not at all. Ultron is engineered to be CPU-friendly. All our models, including DistilBERT and Flan-T5, are optimized to run efficiently on standard consumer hardware without the need for dedicated GPU acceleration.
How fast is the execution loop?
On typical hardware, the entire cycle—from intent classification to tool execution and response generation—takes between 2 to 3 seconds. This provides a near-instantaneous feedback loop for your system commands.
Is it safe to let it execute shell commands?
Security is our priority. Ultron operates with strict user authorization and a whitelist of `allowed_commands`. You have full control over the execution environment, with options for Docker-based sandboxing to ensure your host system remains protected.
Can I integrate my own custom tools?
Absolutely. Ultron is built on a modular "Skill" architecture. You can easily add new Python-driven tools to the `tools.py` registry and map them to new intents in the configuration.