Hey HN,
I’m the creator of Allos. For the past few weeks, I’ve been building a tool to solve a problem that’s been personally frustrating me: every time I wanted to build an AI agent, I felt forced to choose between a simple framework that locked me into one provider (like OpenAI or Anthropic) or a complex one that felt like overkill.
I wanted the ability to switch the LLM “brain” of my agent on the fly without rewriting my code. For example, use GPT-5 for a coding task and then switch to Claude 4.1 Sonnet for a writing task, all with the same agent logic.
That’s why I built Allos. It’s an MIT-licensed Python SDK with a few core goals:
1. True Provider Agnosticism: A unified interface for providers. The MVP has full support for OpenAI and Anthropic. The agent’s code doesn’t change when you switch.
2. A Simple, Powerful CLI: The main way to interact is through a single allos command. You can give it a high-level task, and it will use its tools to get it done (after asking for permission for anything sensitive). E.g., allos “Create a FastAPI app in main.py and run it.”
3. Extensible Tools: It comes with secure, built-in tools for filesystem and shell operations. Adding your own custom tool is just a Python class with a decorator.
4. No Magic: The architecture is simple and transparent. It’s a straightforward agentic loop without a heavy layer of abstraction to fight against. The code is 100% unit tested.
Here’s a 3-minute video showing it in action (including building a multi-file app from a single prompt): https://youtu.be/rWc-8awcAJo
The next big feature on the roadmap is first-class support for local models via Ollama.
I’m here to answer any questions. I’d love to get your feedback, critiques, and ideas.
Thanks for checking it out!
Comments URL: https://news.ycombinator.com/item?id=45858558
Points: 2
# Comments: 0
Source: github.com
