Show HN: Orra – The missing glue layer for production-ready multi-agent apps
github.comWe built Orra after our AI agents crashed in production, costing us thousands in unnecessary LLM calls. It's a Plan Engine that sits between your AI app and execution environment, handling all the messy coordination so you can focus on your agents' core logic.
Key features:
* Works with any language or agent framework (Python, JS today, more coming)
* AI-powered planning with domain safety guardrails
* Production-grade reliability with exactly-once execution
* Run tools as services to reduce latency and hallucinations
Unlike traditional workflow systems, Orra was built specifically for the dynamic nature of AI agents. It's open source (MPL-2.0), self-hostable, and currently in Alpha.
Looking forward to your feedback! Here's the launch blog post for more context: https://outgoing-icecream-a85.notion.site/The-Missing-Glue-L... I don't see local backend in documentation. Is it possible? Hi, yes you can simply run Orra using Docker as a self hosted service locally. Just clone the repo and follow the instruction here: https://github.com/orra-dev/orra?tab=readme-ov-file#2-get-or... Lemme know if you have any issues. I mean, local LLM, for example: can it be run with llamafile instead of cloud AI? Great question! We explored local LLMs (including llamafile-type solutions) in our early development, but found that the reasoning capabilities and consistency weren't quite there yet for our specific needs. That's why we currently optimize for cloud AI models while implementing intelligent plan caching to significantly reduce API costs. This approach gives you the best of both worlds: high-quality execution plans with minimal API costs, plus much faster performance for similar actions. You might find our documentation on plan caching interesting - it explains how we maximize efficiency: https://github.com/orra-dev/orra/blob/main/docs/plan-caching... We're always evaluating new LLM options though, so I'd be curious to hear about your specific use case. Running a 7b coder in laptops with 4060 is possible and with very good results. Orra looks like a very good tool to be integrated with any IDE. Take a look at this: https://github.com/huggingface/llm.nvim -- it has a backend option. Ollama exposes a REST API, I think you guys should support it :) Thanks! Will take a look.