about
One endpoint.
Calibrated probabilities.
That's the whole product.
Updated May 12, 2026
Siftfy is a hosted spam classifier. You POST text to one endpoint and we return a calibrated probability — not a hard label, not a queue, not a webhook to wire up. Drop it in next to your form handler, your comment submit, your sign-up flow, your inbound email — anywhere a human or bot could ship you junk.
Why we built this
Most "spam protection" tools were built when the internet was email and forum comments. Modern abuse comes through APIs — sign-ups, contact forms, listings, marketplace messages, in-app DMs — and the integration story for existing tools is long, lossy, or asynchronous. We wanted a primitive that fit modern code: one HTTP round-trip, calibrated output, no async dance.
Calibrated, not labeled
Every score Siftfy returns is a probability between 0 and 1, not a thresholded yes/no. The model is calibrated — at 0.7, roughly 70% of inputs are actually spam. That gives you a single tunable knob: pick the threshold appropriate to your false-positive tolerance. A help-desk form tolerates more false positives than a marketplace listing; with calibrated scores you can tune both off the same model.
What we believe
- The right amount of magic is none. Predictable inputs, predictable outputs, predictable pricing.
- Latency is a feature. A spam check that takes 600 ms is just an outage with extra steps.
- Probabilistic outputs beat hard labels for any production system.
- Don't store what you don't need. We log metadata; we don't keep request bodies.
Where we're going
Today Siftfy classifies one thing — text spam — and does it well. Next on the roadmap: batch predictions for high-volume jobs, a feedback endpoint so customers can report misses, and adjacent classifiers (phishing, toxicity, PII) accessible from the same calibrated API surface.