Type Safe AI

jevgrep is a command line tool (invoked as jg) that lets coding agents like Claude Code, Codex, and OpenCode find relevant files and source context by asking what the code does in plain English, rather than relying on keyword grep or a prebuilt embedding index. It is powered by Jev, TypeSafe AI's System One model, which scores each code chunk in a single non-autoregressive pass and returns ranked grep-style file:line hits with calibrated confidence . The pitch for this launch is cost: the author reports a 40 percent reduction in coding agent spend on SWE-bench, achieved by offloading repository exploration from a general purpose LLM to a cheaper, faster decision model. The tool is built by David Zhang (@dzhng), co-founder and CEO of Duet (formerly Aomni) , and ships with an installer step (jg skill) that drops an agent skill into whichever coding agent is detected locally so the agent knows when to call jg for context collection instead of reading files into its own window. It runs on Node 22+ and reaches Jev either directly through TypeSafe or via OpenRouter and the Vercel AI Gateway. The underlying model comes from TypeSafe AI, a San Francisco company that exited stealth with $40 million in seed funding led by DCVC and a model designed to put AI directly inside software applications. Its first model, Jev, differs from conventional large language models by producing structured decisions rather than conversational text . TypeSafe was founded in 2024 by CEO Diogo Almeida, who previously worked on RLHF, InstructGPT, ChatGPT and GPT-4 at OpenAI, with co-founders Erik Gafni and Sasha Sheng . jevgrep is one of the first developer tools to show what that model enables when wired into the agentic coding loop that most engineering teams are now standardizing on.