A semantic collision engine, Open Collider addresses the tendency of LLMs to produce predictable, unoriginal outputs by generating non-trivial ideas. It targets researchers and power users, operationalizing Arthur Koestler's bisociation theory to inject distant knowledge domains into prompts. The core benefit is surfacing novel concepts that lie outside the LLM's default output basin, enabling true innovation.

How It Works

The project implements Koestler's principle that creativity stems from colliding incompatible cognitive frames. Open Collider injects principles from structurally distant knowledge domains into LLM prompts, forcing reasoning across low-density regions of idea space. This "semantic collision" generates ideas outside the predictable "Artificial Hivemind," stretching prompts into novel conceptual territory. This approach is advantageous as it systematically moves LLM outputs away from the default basin.

Quick Start & Requirements

Installation involves cloning the repository (git clone https://github.com/CL-ML/open-collider.git), navigating into the directory, and running pip install -e . or pip install -e ".[api]". Python version 3.10 or higher is required. The tool operates in two modes:

• Skill Mode: Free, requires a Claude Code Max subscription. Orchestrated by Claude Code subagents, it is sequential and can be flaky (~25 min/iteration).
• API Mode: Requires an Anthropic API key, uses Python orchestration for parallel LLM calls, is reliable, and costs ~$2-3 per iteration (~10 min/iteration). Setup involves running /collider_setup followed by /brainstorm within Claude Code.

Highlighted Details

Empirical validation across 12 projects demonstrates that Open Collider outputs are systematically further from the default-prompt cloud (semantic embeddings) than baselines like direct prompting, "be original" instructions, or length-matched deep briefs. Blind LLM-judge evaluations confirm Open Collider consistently wins on originality (62-65% win rate) and directionally beats baselines on overall quality (53-59% win rate), proving novelty is not sacrificed for relevance.

Maintenance & Community

The project is developed by Cédric Lion (@oparine_ai), associated with Oparine, a research practice focused on AI creativity. No specific community channels (e.g., Discord, Slack) or public roadmaps are detailed.

Licensing & Compatibility

Open Collider is released under the MIT license, which is permissive and generally compatible with commercial use and closed-source linking.

Limitations & Caveats

The "Skill mode" can be unreliable due to its reliance on subagent coordination. The "API mode" incurs operational costs and requires an Anthropic API key. Furthermore, the "Skill mode" necessitates a Claude Code Max subscription, which may be a barrier to entry for some users.