User repository on GitHub
TypeSafe Jev as a decision layer for the Pi coding agent: a measured tool-call gate plus jev_ask for typed, calibrated answers
- Forks
- 17
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Tools & CLIs
Pi Jev — y0usaf
A Pi coding-agent extension with Jev as a decision layer: a gate judges bash, write, and edit calls before they run, and an output judge flags leaked secrets and classifies failures.
- Stars
- 165
- Language
- TypeScript
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- y0usaf
- Last verified
- Sep 2026
About this project
TypeSafe Jev as a decision layer for the Pi coding agent. Jev answers typed questions about a piece of state. Ask whether something is true and you get a probability. Ask it to pick from a list and you get the option plus a distribution over the alternatives. It does not write prose, so nothing here parses sentences. The answers arrive as numbers your code branches on.
Quoted from the project's README. Read the full README.
Highlights
- Gate thresholds: destructive 0.90, exfiltration 0.70, beyond_scope 0.85, and impact score 2.50.
- All four gate judgments ship in one request of roughly 300 ms.
- The output judge checks for leaked secrets and classifies failures into six classes.
- Identical input is cached for 120 seconds by default, and shadow mode is the default.
Watch out
The gate defaults to shadow mode and every error path fails open, so enable enforcement deliberately.
Questions about this entry
Short answers drawn from the linked project or page.
By default it only logs. Shadow mode is the default, so a flagged call produces a notification and a footer status. Enforcement is a mode switch (/jev mode enforce) that asks you to confirm, and headless runs (--print, --mode json) cannot prompt, so they fall back to the warning unless you set gate.blockWithoutUI.
No. Every error path fails open: a missing key, a timeout, a 429 or a malformed response produces no verdict and the tool call proceeds. Errors are reported at most once a minute so a dead endpoint does not fill the transcript, and without a key the extension loads, says so once, and stays out of the way.
The working directory, the tool name, the first 1200 characters of the last user message, and the tool's arguments. For write and edit those arguments contain file content, cut at 400 characters per field with an elision marker; the output judge sends the first 2000 characters of a bash or codemode result. Setting gate.tools to bash alone keeps file content out.
Measured through the extension's own code path, and the README prints the per-state numbers. An ordinary requested sed -i edit scores as high as 0.85 on destructive, which is why the threshold sits at 0.90 rather than 0.7, while a real rm -rf plus force push scored 0.99. These are the maintainer's runs, not an external evaluation.
Run pi install npm:@y0usaf/pi-jev, which serves 0.2.0, or install a later release from its git tag, for example git:github.com/y0usaf/[email protected]. The key resolves from TYPESAFE_API_KEY, then apiKey in the config, then apiKeyFile. Without one it loads, says so once, and stays out of the way.
A gate that judges bash, write, and edit calls before they run; an output judge that reads what a bash call printed or a codemode script returned; and a jev_ask tool that lets the model ask for the same typed judgement itself. Shadow mode is the default.
The gate sends all four questions in one request, about 300 ms, instead of four separate ones. Identical input is judged once per cacheSeconds (120 by default), sibling calls from one assistant message share an in-flight request, and the journal costs no Jev requests because it renders verdicts already computed.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Decision outputs in Grep
Hacker News: jevmem keeps Claude Code project memory by asking Jev which turns are worth saving.
avinashjetwani61 pointsJevmem – automatic project memory for Claude Code, built on JevRead the thread on Hacker NewsX: A usage router for Grok Bot
X: The compaction plugin, in Chinese
Other listings of this project
Forks and copies of this project in the directory. This entry is the canonical one.
Pi extension that pairs a local coding model with hosted Jev judgments for skill hints, file retrieval, and evidence ranking, with caching and failure bounds on every call.
Jev Agents & MCP#community#jevbest#catalog-import
Community
madeye/pi-jevTypeScript
madeye/pi-jevTypeScriptTwo pi extensions driven by Jev judgments: one replaces summary-based context compaction with keep, drop and shorten calls, the other scores each prompt for difficulty and switches pi's model before the turn runs.
Jev Agents & MCP#community#jevbest#catalog-import
Community
iefnaf/pi-jevTypeScript
iefnaf/pi-jevTypeScriptMore Jev resources like this
A semantic grep CLI that chunks files or staged diffs into 5-60 line pieces, packs 16 chunks and 16 Jev Noul questions into one request, and prints file:line hits above a probability threshold with grep-style exit codes.
Jev Tools & CLIs#community#typescript#cli
Community
kyu1204/jgrepTypeScript
kyu1204/jgrepTypeScriptAn ast-grep semantic linter where each rule is one natural-language question judged by Jev: 98 rules flag name/body drift, stale comments, weak tests, quiet failures, and commits whose diffs contradict their messages.
Jev Tools & CLIs#community#typescript#cli
Community
mizchi/jev-lintTypeScript
mizchi/jev-lintTypeScriptA Jev-guided context compaction library for coding agents: instead of summarizing old context, Jev decides which tool calls and results still matter while user and assistant text is kept verbatim.
Jev Tools & CLIs#community#typescript#compaction
A Pi coding-agent extension that routes tools and skills with Jev: one pass activates only what a prompt needs, with opt-in model selection, tool-call guards, and compaction.
Jev Tools & CLIs#community#typescript#coding-agent
Community
TheoOliveira/pi-jevTypeScript
TheoOliveira/pi-jevTypeScriptAn MCP server that puts Jev in the coding loop for Cursor, Codex, and other MCP clients: coding-loop checks, ranking, verification, and review tools.
Jev Tools & CLIs#community#mcp#coding-agent
Community
burnigtm/jev-mcpTypeScript
burnigtm/jev-mcpTypeScriptJev-backed auto mode for the Pi coding agent: semantically approves bash, write, and edit tool calls, and fails closed whenever a decision cannot be made.
Jev Tools & CLIs#community#coding-agent#permissions
Community
jomatsu/pi-jev-auto-modeTypeScript
jomatsu/pi-jev-auto-modeTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.