About
An engine for telling you no.
The web is full of bias encyclopaedias you browse and a few practices you'd have to remember to do. Almost nothing takes the decision you're actually facing and adversarially audits it. That's the gap Irrational fills.
Why not just ask an LLM?
A frontier model already knows the 188 biases better than any catalogue. So the value isn't informational — it's behavioural. A plain LLM in conversation is trained to agree and help you execute your framing; it rarely, unprompted, tells you you're rationalising. Irrational's engine has a fixed adversarial contract — assume you're biased, refuse to validate, argue the other side — that doesn't drift back to helpfulness. And because it's a tool, it can be invoked systematically (an agent auditing its own draft) rather than only when you remember to.
The evidence backs the stance: research finds current LLMs "passively confirm rather than challenge, reinforcing cognitive bias," and that adversarial questioning is the single most effective debiasing intervention.
The canon
Every entry is grounded in the foundational work: Kahneman & Tversky's heuristics-and-biases programme and prospect theory (Thinking, Fast and Slow); Thaler & Sunstein on choice architecture (Nudge); Dan Ariely (Predictably Irrational — this project's namesake); and Cialdini on influence. Every claim is cited on the sources page.
The shape
Irrational runs no AI model. It provides the curated adversarial framework, the 22-bias catalogue, and a stateless scaffold — delivered as a public field guide and an open MCP tool — and the reasoning is done by whatever model you or your agent already have. No API keys. No stored decisions. Nothing leaves you. That's the privacy story, and the moat: open, composable, and private in a way a closed SaaS can't be.
Open-source — fork it on GitHub.