Tagline Rater
Show HN: A SaaS tagline critic that runs in the browser
Last verified:
What is Tagline Rater?
Tagline Rater is a browser-based AI tool that evaluates SaaS taglines (the hero line under a company's name) and scores them on quality. It distinguishes between sharp, ownable promises that couldn't be pasted on a competitor's site versus forgettable category labels and buzzword filler. The tool runs entirely in your browser on your CPU using transformers.js with a WASM backend, downloaded once and fully offline afterward.
The core rating engine uses intfloat/e5-large-v2 fine-tuned with pairwise ranking loss on hundreds of rubric-rated SaaS taglines plus hand-picked hard negatives like category labels and buzzword soup. It achieves approximately 0.90 held-out pairwise accuracy by fine-tuning the entire embedding body, not just a frozen linear head. Your input line gets embedded (mean-pooled, L2-normalized), then a one-row linear head converts the 1024-dim vector into a score mapped to a percentile against the training set.
The optional brainstorm feature connects to your own OpenRouter key via PKCE OAuth (100% client-side with no app registration or backend) and lets a reasoning model brainstorm taglines. Paste a URL and OpenRouter's agentic web tools fetch and read the page server-side. As the model thinks, candidate lines are mined from the stream and rated live by the local classifier. This provides cloud brainstorming with local curation—no backend on either side.
This tool is designed for SaaS founders, marketers, copywriters, and product teams who need to evaluate whether their hero line is differentiated or just generic buzzwords. The classifier reliably buries imperative formulas like
Tagline Rater pricing
Pricing model: Freemium
Free—completely free tier with no paid plans. The tool runs in your browser with no backend, requires no app registration, and uses your own OpenRouter key for the optional brainstorm feature (you pay OpenRouter directly for their API usage). The model and data are open on Hugging Face.
Tagline Rater pros
- Runs entirely in browser on your CPU—no server required
- Fully offline after initial download
- No backend or app registration needed
- Uses your own OpenRouter key via PKCE OAuth
- 100% client-side privacy for all data
- Approximately 0.90 held-out pairwise accuracy
- Fine-tuned on hundreds of rubric-rated SaaS taglines
- Live rating of brainstormed candidates as model thinks
- Open model and data on Hugging Face
- int8 ONNX export (~336 MB) for efficiency
- Uses transformers.js with WASM backend
- Scores mapped to percentile against training set
- Reliably buries generic imperative formulas
- Surfaces punchy, differentiated taglines
- Agentic web tools fetch and read URLs server-side
Tagline Rater cons
- No built-in tagline generator—requires external OpenRouter key
- Small models can't cheaply be trained to write great taglines
- ~336 MB model download required initially
- Brainstorm feature is optional, not core functionality
- Requires OpenRouter account for brainstorming
- Only works well for SaaS taglines, not other categories
- Trained on few hundred taglines—limited dataset size
- Hard negatives hand-picked rather than comprehensive
Frequently asked questions about Tagline Rater
What is a tagline?
The one line under a company's name—the hero line at the top of a landing page. A good tagline is a tight, ownable promise you couldn't paste on a competitor's site (like 'Wake up an expert'). A bad tagline just names the category or stacks buzzwords (like 'an all-in-one platform for teams').
How does the rating actually work?
The model uses intfloat/e5-large-v2 fine-tuned end-to-end with pairwise ranking loss on a few hundred rubric-rated SaaS taglines plus hand-picked hard negatives. You type a line, it gets embedded (mean-pooled, L2-normalized), a one-row linear head turns the 1024-dim vector into a score, and that score is mapped to a percentile against the training set.
Do I need to create an account?
No account is needed for the rating feature. For the optional brainstorm feature, you connect your own OpenRouter key via PKCE OAuth (100% client-side with no app registration and no backend).
Does my tagline data get sent to a server?
No. The rating model runs entirely in your browser on your CPU using transformers.js with a WASM backend. It's downloaded once and then fully offline. The brainstorm feature is cloud-based but uses your own OpenRouter key with no backend on the app side.
What accuracy does the model achieve?
The held-out pairwise accuracy is approximately 0.90. Fine-tuning the whole embedding body, not just a frozen linear head, is what broke past the ~0.80 ceiling.
Can this generate taglines for me?
The core rating feature doesn't generate taglines. The optional brainstorm feature connects to OpenRouter and lets a reasoning model brainstorm taglines based on your URL, with candidates rated live by the local classifier.
What types of taglines does this work best for?
This is specifically designed for SaaS taglines. The training data consists of rubric-rated SaaS taglines plus hard negatives like category labels, buzzword soup, body-text fragments, and prompt/role leakage.
What does the classifier reliably catch?
The classifier reliably buries imperative formulas like 'Stop X. Start Y.', bare category labels, and literal feature-lists. It floats the punchy, differentiated ones that are ownable and couldn't be pasted on a competitor's site.
Where can I find the model and data?
Both the model and data are open on Hugging Face at standd/tagline-quality-e5-ranker.
Why use a critic instead of a generator?
Great taglines are a taste/creative-leap problem, not a capacity problem—you can't cheaply train a small model to write great ones. A good-enough critic plus a strong brainstormer beats trying to train a great generator. The classifier clears the floor reliably while a reasoning model handles the creative leap.