An idea goes in.
A verdict comes out.
Describe a startup idea in two sentences. Litmus scores its viability, names the companies already doing it, and gives you a 22-week plan with a decision gate you can actually lose.
First review free · about 90 seconds · no card
Viability, 0–100 · model estimate
Held down by one thing: the integrations are public APIs and the narrative is a prompt.
Three steps, ninety seconds.
Describe it
Two or three sentences. Litmus asks for stage, buyer and market only where the answer changes the verdict.
It gets reviewed
The model sizes who pays, finds companies already shipping it, and attacks your weakest assumption rather than flattering the idea.
You get a plan
A score with the reasoning, the competitive landscape, and four phases each ending in a decision that can kill the project early.
It tells you the thing you don't want to hear.
This is the actual output for the idea in the demo above — not a marketing summary of it.
Will a CPA put their name on a narrative a model wrote?
Pricing, liability and the whole automation claim collapse if signing partners insist on rewriting the output. Ask five of them this week. If three say no, you're selling a drafting tool to firms — not a filing co-pilot to founders.
Already doing a version of this
Overlap with your idea · model estimate. Company profiles are qualitative — verify before planning around them.
Compare it to what you'd do instead.
The alternative isn't another tool — it's three weeks of your own time, or an advisor's day rate.
Asking around
- People who know your market
- Everyone is too polite to say it's derivative
- No competitive scan
- Nothing written down to argue with later
Litmus Pro
- Unlimited reviews, 90 seconds each
- Scored against five weighted axes, with reasoning
- Named competitors and where you'd win
- A phased plan you can hand to a cofounder
Free plan: one review a month, full report.
Reasonable objections
Is a model actually qualified to judge my idea?
No — and Litmus doesn't claim to be. It's fast, structured pushback: it surfaces who already does this, what the weakest assumption is, and what it would cost to test. The judgement stays yours. Every number is labelled as an estimate.
What stops it just telling me the idea is great?
The review is built to lead with the reason it fails. Every report ends on an unresolved question you have to go answer offline, and the score is held down by the weakest axis rather than averaged upward.
Do you train on my ideas?
No. Submitted ideas stay in your account and are never used as training data. Reports are kept for 12 months so you can compare a re-run against the original.
Are the competitors real?
The companies named are real and the profiles are qualitative — we don't publish funding or revenue figures we can't stand behind. Overlap percentages are model estimates, labelled as such.