The AI market has seen a rapid influx of decision-making models following the release of Jev by TypeSafe. These models provide direct probability-based answers to specific queries, avoiding the token-generation processes and hallucinations associated with traditional large language models.
Nandakishor Mukkunnoth, founder of ConvAI Innovations, claims the architecture mirrors his own research. He published his SalesRLAgent model weights and dataset on March 30, 2025, and followed up with a second paper in September. Mukkunnoth subsequently released Laya, an open-source alternative that runs on local hardware.
Major industry players have launched their own versions of the technology within weeks. These include Microsoft Decision, OpenAI Decisions API, and Fastino Glide. Additionally, Cloudflare Clef and Perplexity Decider were released with open weights for local deployment.
The open-source ecosystem has also expanded with projects such as Amazon's Strands Decider 2B, Kev, Von, Jeff, and NanoJev. Most commercial offerings are built on variants of Alibaba's Qwen model, while Mukkunnoth's Laya utilizes encoder-only models from the Bert family.
