Models are getting smarter.

Models are getting smarter. I care about what they’re allowed to do next. I’m building Nika, an open-source governed runtime for AI workflows. You describe a workflow in .nika.yaml, Nika checks what it’s allowed to do before execution, runs models/tools/agents, can stop for human approval, and keeps Proof of what actually happened. Intent → Check → Run → Proof What I want to explore with Jev is: Jev → typed probabilistic evidence → deterministic policy → authority → action / human → Proof So Jev can make thousands of cheap bounded decisions, while Nika keeps control over what those decisions can actually trigger. Long term: Jev for decisions, Claude/GPT for reasoning, agents for open-ended work, deterministic code for effects. Same workflow and authority model around all of them. I’m building the Jev integration + a benchmark against normal LLM structured outputs. OSS: https://github.com/supernovae-st/nika would love