Shashank-H/pi-jev-context-curator
User repository on GitHub
A Jev based context curator for pi
- Forks
- 2
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Agents & MCP
Pi Jev Context Curator
Pi extension that judges context units with Jev before each model call, keeping what a future response needs and omitting the rest without editing the stored session.
- Stars
- 7
- Language
- TypeScript
- License
- MIT
- Published by
- Community
- Author
- Shashank-H
- Last verified
- Oct 2026
About this project
A pi extension that curates the context sent to the LLM before each call — using TypeSafe AI's Jev model to decide which context is actually needed. Before every LLM call, pi fires the context event with the full message list. This extension: Groups messages into units — an assistant tool-call message plus its tool results stay together as one atomic unit, so curation never orphans a tool result.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- Groups assistant tool calls with results into atomic units so curation never orphans a tool result.
- Records judged units by content fingerprint and sends only new units after the checkpoint to /v1/systemone.
- Uses one authoritative noul keep and remove plus two choice probes that only explain, never override.
- Writes removed units as expandable [jev removed] entries with a popup; records stay display-only.
Watch out
Fail-open by design: missing keys, errors, timeouts, and rate limits pass context through untouched, while system messages, the latest unit, summaries, and sub-threshold contexts are always kept. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
Nothing is deleted. Pi only lets extensions transform the per-call payload, not the stored session, and tool calls stay grouped with their results as atomic units. The checkpoint is cleared on session start so judgments never leak across sessions.
The primary noul controls keep and remove while the two choice probes only explain and never override it. Missing or non-numeric primary answers are kept, and the threshold defaults to 0.25 where lower keeps more context.
Each unit is judged once with usually 1-3 new units per turn, so cost stays tiny as sessions grow. Jev answers in about 70-500 ms at $42 per billion input tokens with free outputs, and curation runs only above 8000 estimated tokens by default.
Install with pi install from npm or git, or drop the folder into the pi extensions directory. Set the key with the set-key command for file mode 0600 storage, or use TYPESAFE_API_KEY; clear-key removes it.
Every unit Jev has already judged is recorded by content fingerprint, so past decisions are reused and the same data is never sent to Jev twice. Only units after the checkpoint are sent, and because each unit is judged once the per-turn cost stays tiny regardless of session length. The checkpoint clears on session_start, so judgments never leak across sessions.
CURATOR_JEV_THRESHOLD defaults to 0.25, keeping a unit when Jev's needed probability is at or above it, and lower is more conservative. CURATOR_JEV_MIN_TOKENS defaults to 8000 so small contexts pay no extra latency, CURATOR_JEV_FREQUENCY defaults to 5 so Jev is queried every fifth call, and CURATOR_JEV_ENABLED set to 0 starts it disabled.
/jev-context-curator status shows state, settings, key source, checkpoint size, session savings, and session cost; stats details removals and cost; inspect opens a dashboard of kept and removed units with Jev's reason. on, off, set-key, clear-key, threshold, min-tokens, frequency, and reset round out the command set.
More Jev resources like this
20GitHub stars
TypeScript SDK and CLI coding agent where Jev only picks options while your code runs effects, building programs as syntax trees.
Jev Agents & MCP#community#awesomejev#catalog-import
Community
rhighs/jev-codeTypeScript
rhighs/jev-codeTypeScript22GitHub stars
Runnable agent harness that moves pick, score and yes-no loop choices to Jev or Laya in milliseconds while reserving the language model for writing, compiling DECISION.md into executable questions.
Jev Agents & MCP#community#jevbest#catalog-import
Community
zjunlp/JevLoopTypeScript
zjunlp/JevLoopTypeScript35GitHub stars
Diff-aware test selector that sends the git diff to Jev with one question per test, then prints the native filter flags and hands them to vitest, jest, node, bun, Playwright, cargo, or go test.
Jev Agents & MCP#community#jevbest#catalog-import
Community
mizchi/jev-test-filterTypeScript
mizchi/jev-test-filterTypeScript13GitHub stars
Pi extension that classifies each task with Jev over OpenRouter and routes it to a Pareto-optimal model balancing measured quality, cost, and latency under a role policy.
Jev Agents & MCP#community#jevbest#catalog-import
15GitHub stars
Node CLI that lints a file tree against fuzzy natural-language rules in one Markdown file, judging each matched file with a swappable TypeSafe-backed judge.
Jev Agents & MCP#community#jevbest#catalog-import
Community
tyler-dot-earth/patdownTypeScript
tyler-dot-earth/patdownTypeScript106GitHub stars
Decision layer exposing Jev's typed yes-no, choice, and score answers to DeepSeek Harness and MCP hosts, offline by default with every network transmission disclosed per feature.
Jev Agents & MCP#community#jevbest#catalog-import
Community
PerryLink/jevcoreTypeScript
PerryLink/jevcoreTypeScriptFrom the Jev community
Posts from builders shipping with Jev right now.