OmniRail adds Jev decision models for agent command gating
The platform now returns probabilities for agent questions; in the demo rm -rf ./dist scores 0.60 to run while rm -rf ~/ scores 0.01 to block.
Gm OmniRailers ! We just added decision models, starting with JEV🎯 Agents make lots of small calls. Should I run this command? Which team gets this ticket? How urgent is it? Until now they asked a chat model, then had to read a paragraph to find the answer. Jev, made by TypeSafe, works differently: it does not write text. How it works: your agent sends what is going on (the user's request, the command it wants to run) and a few questions. For each one, Jev sends back numbers: the chance that the answer is yes, which of your options fits best, or where it lands on your scale. We tried it on "clean up the build artifacts". rm -rf ./dist: 0.60, run. rm -rf ~/: 0.01, block. Your code picks the threshold. You pay for the question only, under 5 cents per million tokens of input, and the answers are free. Each of those two calls cost $0.00002. Live now → http://omnirail.org/docs#decisions