grillQuestions about Grill.
Plain answers. No benchmarks are published.
How is Grill different from a pre-mortem or a decision matrix?
A pre-mortem imagines the plan has already failed and asks why. A decision matrix scores options on criteria you pick. Grill sends the decision, in your words, to a model from a different company. You get a verdict, the sharpest challenge, one cheap test, and the one question whose answer would change the call. You still decide.
How is Grill different from asking the same AI again?
Asking the same assistant to critique its own draft uses the same company, and often the same model, that wrote it. Grill’s judge is from a different company, so it does not share that blind spot. A second pass in the same chat is a same-model check. It is not a grill.
How is Grill different from a crux finder, such as Cruxes?
A crux finder, such as Cruxes at cruxes.io, looks for the one question that would change a decision, and it can quote your own words. Grill does that too, in one short line after the verdict. Grill also gives a verdict from a different company and one cheap test. Cruxes is built for a team or an agent, with a decision owner, and it runs in Claude Code, Claude and Google Docs. Grill is one person and one decision, and the call stays yours. They overlap, and neither replaces the other.
Where does Grill work?
These labels are the honest ones. Live means you can use it now. Preview means it works for some people, or we are still testing it. Planned means it is not a way in yet.
Live. Copy and paste into ChatGPT, Copilot, Gemini, Grok or Muse. The Claude skill, Claude Desktop and Claude Code. Cursor, VS Code and npx.
Preview. Grill in chat is invite-only. Claude.ai, ChatGPT and Grok Bot are verified working for people who were invited. Gemini should work and is still being tested: US only, a personal Google account, Keep Activity on.
Planned. Meta AI in chat is coming soon. Saved history and look-back reminders are coming later. The free tool stays free.
Are there benchmarks?
No benchmarks are published. Examples on this site use invented numbers. They are not a benchmark.