pi-jev-context
Pi extension that dedupes repeated file reads and, opt-in, has Jev mark passing-test noise in long logs, with verbatim recall.
Overview
pi-jev-context (Nyarlathoteppppp) keeps repeated reads and noisy logs out of Pi's context without losing them. It has two independent layers and a recall tool. Dedupe, on by default and needing no network, turns lines you were just shown, unchanged and in the same position, into a one-line marker when a file is read again. Sieve, opt-in, sends long test, build or lint logs to Jev to mark passing-test noise, while code keeps every failure, warning, stack trace and summary. A context_recall tool returns any shortened output verbatim by id or keyword.
Only arriving output is shortened; earlier messages are never rewritten, so the cached prefix stays intact. The README is candid: replaying 24 real sessions, default dedupe saved only 0.2% of tool-result tokens, and the sieve hid about 11.8k tokens on two synthetic tasks that still passed, described as a smoke test. A /context report command shows effects in your sessions. It works with a TypeSafe or OpenRouter key.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this block of log output passing-test noise? | yes · no |
How it works
- Dedupe repeated file-read lines locally.
- With sieve on, send long logs to Jev to mark passing noise.
- Keep all failures, warnings, stack traces and summaries in code.
- Recall shortened output verbatim when needed.
Strengths
- Cache-neutral; earlier messages never rewritten.
- Lossless via context_recall.
- Honest, measured savings.
Caveats
- Dedupe savings are small in real sessions.
- Sieve sends logs to Jev and is opt-in.
Getting started
pi install git:github.com/Nyarlathoteppppp/pi-jev-context, then /context sieve on with a TypeSafe or OpenRouter key.At a glance
- Type
- Benchmark
- Status
- beta
- Language
- TypeScript
- Runtime
- Pi extension
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Model performance first. Token savings second. A Pi extension with freshness-aware read dedupe, Jev log filtering, and searchable verbatim recall. Keeps existing message history intact.
logicrw/awesome-jev-projectsCache-neutral context trimming for the pi coding agent, powered by TypeSafe Jev: long tool output cut to verbatim key lines before it enters context, with lossless recall. Measured, with pre-registered benchmarks.
hellogumbo/awesome-jevPi extension that collapses repeated file reads and, opt-in, has Jev mark the passing-test noise in long logs while code keeps every failure; a recall tool returns anything shortened verbatim.
MrJev/awesome-jev
Frequently asked questions
Does dedupe need a key?
No, it runs locally.
What does the sieve keep?
Every failure, warning, stack trace and summary; Jev only marks passing-test noise.
Can the model get trimmed output back?
Yes, through the context_recall tool.
Related projects
Prunes long command output for coding agents by asking Jev which lines matter for the task, keeping original text and order.
ProjectbetaTypeScriptCodex plugin that builds handoff packets by asking Jev whether each old read-only tool record should stay, keeping retained text verbatim.
ProjectbetaJavaScriptAgent plugin that asks Jev whether the session is at a safe boundary to /compact and can trigger it on Pi and Claude Code.
PluginbetaTypeScriptPi extension where Jev judges each completed tool call and result before compaction, keeping key evidence verbatim in the summary.
PluginbetaTypeScriptOh My Pi extension that truncates tool calls and results Jev scores as no longer needed, while keeping all user and assistant text verbatim.
PluginbetaTypeScriptPi extension that prunes stale tool calls with two Jev yes/no questions each, keeping text verbatim and leaving summaries to Pi.
PluginbetaTypeScript