Core Features
Designed for directness and power, with minimal abstraction.
Direct Shell Access
Connect LLMs directly to your system's shell and filesystem. It's the most straightforward way to give your agent real-world capabilities.
Advanced Orchestration
Leverages recursive decomposition, planning, and multi-agent workflows to tackle complex, multi-step tasks automatically.
Model Agnostic
No vendor lock-in. Our pluggable architecture allows you to easily use any model from any provider that fits your needs.
Fully Hackable
Easily customize and extend everything. Write your own agents, tweak system prompts, or define new actions to match your exact workflow.
Training-Ready Data
All agent events and interactions are stored as local files, providing clean trajectory data perfect for fine-tuning your own models.
CLI & Web Interfaces
Engage with your agents through a powerful command-line interface or a user-friendly web UI. Your choice.
Quick Start
Get up and running in seconds with a simple pip install.
Join our open source community and help shape the future of AI agent development.
Our Philosophy
Guiding principles for building agents that are powerful, transparent, and built for the long run.
Less is More
We focus on giving the LLM a few, powerful actions. This enables maximum capability with minimal human intervention in the workflow.
You Are in Control
We don't enforce permissions. Run agents with models you trust, guide behavior with prompts, and use a VM or container for sandboxing.
Designed for Optimization
Minimal abstraction means you can easily dig into prompts and responses to tune agent accuracy—critical for moving beyond demos.
The Goal is Self-Iteration
The ultimate vision. Anges is built with its own help, and we believe the future is agents that can improve and iterate on themselves.
See It in Action
Explore these real-world examples of Anges tackling development tasks.
Linux and Cloud Ops
Interact with the runtime shell and manage cloud resources using the CLI.
Creating a Demo Website
Build a file upload and Q&A Flask website using a single prompt.
Developing This Landing Page
Iterate the landing page application to integrate a Google Interest Form.
Writing Features with Tests
Demonstrate Anges iterating its own codebase to implement requested features with automated testing.
Solving a SWE Bench Problem
Show the workflow of Anges tackling an issue in SWE Bench Verified.
Recursive Task Handling
Exhibit recursive self-calling capabilities for resolving complex tasks with sub-tasking.
Join the Community
Connect with developers building the future of AI automation.