Context compaction
Decide which tool calls and results still matter so stale context can be dropped.
38 rows are filed under this pattern, 38 of them with code. Rows are listed as the catalogue orders them: official first, then rows with code, then by stars.
Filter and search them in the catalogue →
Rows marked “not Jev itself” or “shadow mode” stay listed, but they are not examples of deciding with Jev.
-
Hermes Agent: Jev compaction evaluation
Ported the Jev compaction approach, measured it against their shipping summariser, and published the conclusion not to adopt it.
-
jcode: memory recall without embeddings
Replaces the whole retrieval stack for memory recall — no embeddings, no BM25, no reranker — with one batched Noul per candidate memory.
-
fast-jev-compaction
A Claude Code plugin that replaces the compaction summary with per-item decisions: stale tool calls are dropped or truncated, everything kept stays verbatim.
-
jevgrep (dzhng)
CLI and agent skill that finds code by asking what it does: `jg "How are telemetry events recorded?" ./repo` has Jev judge relevance across folders, files and declarations and returns files and verbatim excerpts. The author reports the same 8 of 10 SWE-bench tasks solved at ~30% lower cost.written by a model
-
hermes-jev-skills
Nine agent skills plus a CLI covering model routing, memory filtering, turn retention, one-of-many skill selection and next-action choice.
-
mu
Coding agent and desktop app built on pi that asks Jev at 38 decision points which tool-output chunks enter the context, which stale tool results to drop, whether a rule-flagged command was asked for and whether fetched pages or MCP output carry injected instructions.
-
compact-adviser
Plugin answering one question, should I /compact now: one Jev request asks two choices (is the latest unit of work finished; hands-on or coordination), combined in code into a score whose floor relaxes from 0.90 to 0.50 as the context fills. It hints, or compacts on Pi and Claude Code if you opt in.written by a model
-
jev-pruner
Trims long shell output before the model sees it, asking one Noul per chunk.
-
Quicksilver
Claude Code skill that hands relevance calls to Jev: one noul per file, log line or ticket ('which of these files handle auth?'), asked in parallel and sorted by probability, so Claude gets a shortlist with confidences instead of reading everything.written by a model
-
Winnow
Context garbage collection for Claude Code: when Read, Bash or Grep dump a wall of output, each chunk is judged for relevance to the current task.
-
save-token-jev-clean
Portable, Jev-guided context compaction for coding agents: instead of an LLM rewriting old context into a lossy summary, Jev decides which tool calls and results still matter, and user and assistant text is kept verbatim.
-
yoshi
Context-pruning proxy for Claude Code and Codex: Jev judges which history is still needed, measured not claimed. POC here now, heading soon into https://github.com/compozy/compozyupstream description
-
claude-jev
Claude Code and Pi plugin that sends small judgments to Jev: whether an edit breaks a rule in CLAUDE.md (blocked with a file:line citation), which context blocks to keep verbatim at compaction, a model tier per subagent with nouls on its brief, and a routing hint per prompt. Every hook fails open.written by a model
-
lcc
Local Context Compiler (lcc): clean, dedupe and compact prompt context before it reaches the model, then report every block dropped, the cache tokens a pass invalidates and when pruning pays off. Runs offline with a local 1K decision model. MIT, no API key, zero telemetry.earlier upstream description
-
dsh-jev-tools
DeepSeek Harness plugin asking Jev, before tool output enters the context, which segments of a long read or fetch matter (kept by rank, not threshold), whether a fetched page holds instructions aimed at an AI (warned at 0.75, left unchanged), which skill to suggest, and if a done claim has evidence.written by a model
-
omp-jev-compaction
Verbatim Jev-scored context reduction for omp, over TypeSafe or OpenRouterupstream description
-
fast-dev-compaction
Codex plugin: verbatim Jev-guided context restoration around session compaction. Port of tamaratran/fast-jev-compaction to Codex lifecycle hooks.upstream description
-
jev-compact
Jev-scored context compaction for OpenAI Codex CLI — scores every tool call before compaction and restores critical tool outputs verbatim after itupstream description
-
pi-jev
Pi extension suite powered by Jev: selective context compaction and model routingupstream description
-
pi-jev-context
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.upstream description
-
jit-context
Architectural Moat: JIT-JEV Context OS — Epistemic runtime & JEV System 1 context gate for AI agents (L0 SQLite WAL <3ms, Epistemic Invariants I1–I10, CERN Zenodo DOI: 10.5281/zenodo.22649542)upstream description
-
dsh-jev-prune
Jev-judged context compaction for DeepSeek Harness: semantic tool-result pruning + deterministic receipt compactionupstream description
-
jev-docs
Community-maintained history of Jev / TypeSafe System One APIs, SDKs, agent guidance, and engineering best practices.upstream description
-
pi-jev-context-curator
A Jev based context curator for piupstream description
-
deepseek-harness-jev-pre-compaction
A pre-compaction advisor for DeepSeek Harness. Runs before the standard `compaction-basic` backend, using TypeSafe JEV to safely prune low-value tool results from model context. Original session events stay in the append-only log; only the model-visible view is replaced with compact markers or
-
fast-compaction-dsh
Verdict-based context compaction for DeepSeek Harness — replaces lossy LLM summaries with fast keep/truncate/drop decisions from jev-latest; everything kept stays verbatim. Port of tamaratran/fast-jev-compaction.upstream description
-
jev-compaction
A context compactor that can only score, never write — so an agent's memory can't hold a fact the transcript never contained. Working demo, runs offline.upstream description
-
pi-jev-compaction
Automatic Jev context clearing for Pi. Keep the conversation, prune stale tool output, retrieve originals without rerunning commands.upstream description
-
Jev by Example
Ten runnable JavaScript agent decisions, one file each: reconciling a new memory against a stored one, gating whether an HTTP 200 really satisfied the task, retry vs. reconcile after an uncertain write, scoring context against a budget, checking a handoff for dropped prohibitions.
-
jev-compaction
Context compaction for Hermes sessions that never summarises: every tool call is scored by TypeSafe's Jev model, stale calls are dropped, everything kept stays verbatim.upstream description
-
jselect
Useful evidence for your AI, within a token budget. A fast, source-linked context selector for files, records, and agents.upstream description
-
pi-fast-jev-compaction
Verbatim context pruning for the pi coding agent, scored by TypeSafe Jev: stale tool calls and results are dropped or truncated, everything kept stays verbatim.upstream description
-
jevprune
Filter command output for coding agents using a task description.upstream description
-
pi-fast-jev-compaction
Fast JEV compaction extension for piupstream description
-
pi-jev-context
Reversible context pruning for Pi, powered by TypeSafe Jev. Keep useful context without deleting session history.upstream description
-
deepseek-harness-jev (luobosibing2)
DeepSeek Harness plugin with twelve opt-in features using Jev as the judgment model: skill selection and glob-result ranking (one `noul` per candidate), completion and goal checks, drift reminders, interjection routing, tool-output and test-log filtering, and approval of single workspace operations.written by a model
-
pi-jev-compact
Selective, verbatim context compaction for Pi: Jev scores every tool call and result, unneeded calls are dropped and the rest stays verbatim, with no LLM-written summary.
-
smoking-extraction-benchmark
Synthetic smoking-history extraction benchmark comparing TypeSafe Jev and OpenAI structured outputs, with reproducible accuracy, cost, and latency results.upstream description