What it is
Oko is a local code search tool for coding agents like Codex, Claude Code, and OpenCode. It shortlists function-level chunks using ripgrep and BM25, then ranks them with Jev to return relevant excerpts via MCP, reducing tokens and tool calls.
How it uses Jev
For each shortlisted chunk, Oko asks Jev a Noul relevance question across three parallel requests. The accepted chunks are returned as excerpts through MCP. The cutoff and excerpt selection are implemented in code.
Primitives:noul
Technique worth stealing
Combine ripgrep and BM25 for shortlisting, then use Jev's Noul relevance to rank chunks.
Try it
Run `oko ask "where is authentication handled?"` in your project directory.
Evidence
Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.
- Jev-centricyes0.80
- Shows a System One patternyes0.91
- Handles uncertaintyno0.08
- Measuredyes0.96
- Runnableyes0.81
- Worth recommendingno0.29
- Model replicano0.06
- Problem scopescore on a 0–2 scale1.52
- About Jevyes0.96
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 23 Sept 2026.