The problem
I work on many projects every day. Some of them are experimental and exciting, and there are always competitions or similarities in other projects. But the main problem I face is reinventing the wheel.
You often find that the solution you've been looking for is already out there. The more problematic case is when it isn't in the field you're working in, but in a distant one, where it's already mature and battle-tested.
The idea
I was working on a personal harness project for AI agents. Right around the time I found a solution for what I like to call AIX (the experience of the agents with their tools and integrations), it came to mind that I could be wasting time.
The solution might already exist in another field of study. So why not create a tool that looks into other fields? Their definitions, their research papers, their connections and more.
The flaws
At first, I hit a problem. The agents I work with rejected the tool's output and called it noise. I had to explain it to them by giving the right framing: what the problem is, and what solution I'm offering. And here, it worked.
The agents started coming up with not just one idea to solve the problem, but many, several steps ahead, like a game of chess. They found what they needed in each connection the tool found.
The surprise
I was even able to make them develop the tool itself, by asking them to find solutions related to the compound effect.
Beyond code
I tested it on many other things, like languages and art. Each time, the tool kept showing me new use cases I never thought it would work for. It feels like some healthy version of ADHD but for agents.
LLMs work by retrieving similar information, but the tool works to make them think beyond the familiar just like thinking outside the box.
It still depends on how capable the model itself is and other integrations can either sabotage or push that even further.
Try it
The tool's name is Eonrise.
It was open-sourced just recently at
github.com/Superstos/Eonrise, and it's also available as a Python package:
pip install eonrise
For your agents
Install it as a skill, so AI agents know when to run it and how to read its output:
Claude Code
/plugin marketplace add Superstos/Eonrise
/plugin install eonrise@eonrise
Codex
codex plugin marketplace add https://github.com/Superstos/Eonrise
codex plugin add eonrise@eonrise
Top comments (1)
superstes, this is a fascinating approach to breaking llm myopia. the observation that agents initially reject cross-domain data as "noise" until given the right framing is a profound insight into attention mechanisms and prompt engineering.
llms are heavily biased toward semantic similarity. forcing them to evaluate distant domains requires explicit cognitive scaffolding as a search engine, but as a "lateral thinking" bridge, you are essentially hacking the agent's reasoning loop to evaluate analogical solutions rather than just direct matches.
this is essentially "lateral rag" (retrieval-augmented generation). instead of just fetching the most similar document, you are deliberately injecting controlled, framed novelty to spark multi-step, chess-like reasoning.
fantastic open-source contribution. building tools that make agents deliberately "think outside the box" is exactly how we move from basic task automation to true agentic problem-solving. 🐯🧠