Nyckel

Nyckel is a hosted machine learning platform for building classifiers and detectors from labeled examples and calling them through an API. A team uploads samples, Nyckel trains a model automatically, and the model returns predictions with confidence scores. Feedback from real predictions can be fed back in so the model improves over time. It works on images, text and structured data. Image uses include defect detection, species identification, content moderation, product categorization and object detection and counting, along with OCR and similarity search. Text uses include spam detection, ticket routing, sentiment analysis and intent classification, and structured data uses include fraud detection, churn prediction, anomaly detection and lead scoring. The vendor also offers ready made classifiers and detectors that can be tried on a user's own data, and positions the platform as a source of dependable decisions next to generative models, which it says are better at writing text than at making repeatable choices. Developers are the main audience, with a quickstart that reaches a first prediction in five API calls, a full API reference and guides. Pricing is published and has a free tier, two paid tiers and a custom one. Free costs $0 per month and includes 100 invokes, up to 200 samples and up to 5 functions, with $0.005 per invoke beyond that. Starter costs $149 per month and includes 15,000 invokes, up to 30,000 samples and up to 20 functions, with $0.005 per extra invoke and $0.0025 per extra sample. Business costs $599 per month and includes 300,000 invokes, up to 600,000 samples and up to 100 functions, with $0.001 per extra invoke and $0.0005 per extra sample. Enterprise pricing is arranged with the vendor.