CL-ML

CL-ML / open-collider

Public

A semantic collision engine for non-trivial LLM idea generation. Operationalizes Koestler's bisociation theory (1964): injects structurally distant knowledge domains into the prompt, forces collisions, surfaces non-trivial ideas. Empirical validation in CL-ML/open-collider-research.

34
2
89% credibility
Found May 19, 2026 at 188 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Open Collider is a creative ideation engine that generates genuinely original ideas by forcing AI to reason through concepts from completely unrelated fields before generating output. Instead of asking an AI directly (where responses converge to predictable patterns), it injects material from distant domains like glass physics, fermentation biology, or supply chains—creating 'collisions' that produce ideas that couldn't exist without that structural tension. The tool runs brainstorming sessions with multiple iterations: generating distant domains, colliding them with your reference materials to produce hundreds of raw ideas, scoring those ideas on five quality dimensions, and letting you curate with love/like/trash feedback. That feedback then guides the system to either deepen productive territory, explore fresh domains, or refresh successful patterns into new disciplines. Built by Oparine research, it includes a 12-project benchmark proving its outputs sit 4-13x further from the 'default AI response' cloud than standard prompting, with blind judges preferring its ideas on originality.

How It Works

1
đź’ˇ You have a creative challenge

You need fresh, original ideas for a project—maybe redesigning a product, solving a business problem, or finding a new angle on something that matters to you.

2
🛠️ You set up the tool

You install Open Collider and connect it to an AI service. Then you run a simple setup command where you describe your creative brief and share any reference materials that capture your problem space.

3
🔀 The magic happens: distant ideas collide

The tool pulls knowledge from completely unrelated fields—glass physics, fermentation biology, supply chains—and forces your brief to collide with these alien concepts. This is where the magic lives.

4
đź“‹ Hundreds of ideas flood in

The system generates hundreds of raw ideas by combining your reference materials with these distant domains. Most are noise, but buried in the pile are genuine gems.

5
Choose how you want to run it
🚀
API mode

Your computer talks directly to the AI. Everything runs in parallel—faster but requires your own account.

đź’Ž
Skill mode

Claude Code Max orchestrates everything as subagents. Slower but free with your subscription.

6
🎯 You curate and give feedback

The system scores every idea on five dimensions and shows you the best ones. You mark each one as loved, liked, or trashed. This feedback shapes the next round.

7
🔄 The system gets smarter

Loved ideas push the system to dig deeper into productive domains. Fresh domains keep exploring new territory. After 3-5 rounds, the session naturally exhausts.

✨ You walk away with a curated idea report

The tool compiles everything into a beautiful report: your best ideas organized by iteration, flagged by your feedback, and ready to share or pursue.

Sign up to see the full architecture

6 more

Sign Up Free

Star Growth

See how this repo grew from 188 to 34 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is open-collider?

Open Collider is a Python tool that forces LLMs to generate genuinely novel ideas by crashing distant knowledge domains together inside the prompt. Instead of asking an AI directly (which produces the same predictable outputs every time), it injects material from fields the model would never reach on its own—glass physics, supply chains, fermentation biology—then lets the collision produce ideas that couldn't exist without that friction. It runs as a Claude Code extension with two modes: a free Skill mode for Max subscribers, and a faster API mode requiring an Anthropic key. You define your brief, run brainstorm iterations, and the system generates hundreds of ideas, scores them on five axes, and surfaces the ones worth your attention.

Why is it gaining traction?

The hook is the problem nobody talks about: ask an LLM for ideas twice and 80% of outputs cluster in the same region. Different words, same substance. This tool attacks that directly with a theory-backed method (Koestler's bisociation from 1964) and shows empirical evidence—12-project benchmarks where outputs measurably escape the default-prompt basin and win blind preference tests against baselines. The three-strategy system (Fresh explores, Deepen exploits productive territory, Refresh transfers what works) gives it an adaptive quality that pure prompting lacks.

Who should use this?

Product strategists and R&D teams hitting a wall with conventional brainstorming. Innovation leads who need non-obvious ideas for complex problems. Researchers studying LLM creativity limitations. Anyone willing to invest setup time (configuring briefs, reference texts, scoring axes) in exchange for outputs that actually surprise them. Not for quick one-off requests—this is a session-based workflow.

Verdict

The theory is solid, the empirical backing is rare for a project this small, and the output quality difference is real. But with 34 stars and a 0.9% credibility score, this is experimental software with minimal community validation. Try it if you're working on genuinely hard ideation problems and have Claude Code Max or an Anthropic key; otherwise, wait for the research repo to mature. The method works—the tooling needs more eyes.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.