Video
CommunityJev Tools & CLIs
Fast Jev Compaction Demo
Tamara's demo of fast-jev-compaction for Claude Code: instead of summarizing context, Jev scores every tool call and drops the irrelevant ones, with a video of it running.
- Format
- Video
- Reading time
- 2 min
- Published by
- Community
- Author
- tamara
- Last verified
- Sep 2026
Works forJev for Tool calling
Highlights
- Replaces summarization-based compaction with per-tool-call keep or drop decisions.
- The post calls it instant compared with a summarization pass.
- Video shows the plugin running inside a Claude Code session.
Watch out
A promo post from the plugin's author; no independent benchmark.
Questions about this entry
Short answers drawn from the linked project or page.
It never writes prose. The idea is to score each tool call and drop what is judged irrelevant, so nothing is summarized and no text is generated, and the linked repo describes itself as replacing the compaction summary outright. The post does not say what happens to conversation text between those calls.
It is lossy by design. A dropped call may be the one holding a file path, a decision, or a failed attempt the model needs later, and the post offers no recall measurement. Diogo Almeida replied endorsing it as free coding agents from designing around the KV cache, which is a claim rather than a benchmark.
Undocumented beyond the plugin existing. It runs as a Claude Code plugin and the repo is linked from the author's next-day post, but the announcement says nothing about whether history is rewritten in place, whether any summary is still produced, or how to undo a prune you did not want.
No independent benchmark exists yet; this is the plugin author's own promo post. There is no cost figure, no before-and-after token count, and no measured task success rate in the announcement, so the word instant is a qualitative claim about a summarization prompt not being involved.
tamara, posting as @tamarajtran. Her bio says she built three number-one social apps with twenty million users and previously worked at Apple, which is a consumer-social background rather than a context-management one. That is context for judging a claim about how coding agents should compact history.
About 3.9 million views, with the Jev founder replying to endorse it. Reach like that shows the idea resonates with people frustrated by summarization-based compaction, but it is not a measurement: there is still no token count, no loop rate and no task-success number anywhere in the thread.
In tamaratran's GitHub repository, fast-jev-compaction, described as a Claude Code plugin that replaces the compaction summary. The post itself gives no install command, since a separate post by another user supplies the paste-this-prompt steps, so the repository is the place to check setup and permissions.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: The compaction plugin, in Chinese
X: The compaction plugin, in Spanish
From 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.