Context compaction stopped being a summary and became a checklist. fast-jev-compaction pairs every…
context compaction stopped being a summary and became a checklist. fast-jev-compaction pairs every tool call with its result and lets jev vote keep or keep-verbatim, so file paths, exact errors and failed attempts survive instead of being rewritten away (sept 20).
what an agent forgets is now a decision, not a rewrite.
Context
The reference doc for the pi-fast-jev-compaction 0.2.0 extension says every tool call is matched with its result by id, Jev answers questions per call and result, and a threshold with a default of 0.5 decides keep, truncate or drop; it says Jev compaction never rewrites text and falls back to pi's default summary when Jev fails or would free too little. TypeSafe's page says that instead of asking a model to summarize history you ask Jev a yes/no question about each old item, at roughly 70 to 500 ms per decision. A ClaudeWorkshop article dated 22 September 2026, which the site labels as AI-assisted drafting and not a hands-on test, describes a Claude Code plugin by tamaratran.
The docs describe keep, truncate or drop with a confidence threshold, not a two-way vote of keep or keep-verbatim. Original messages and kept tool results stay verbatim while dropped or truncated outputs are removed, so failed attempts survive only if Jev judges them relevant, and no recall measurement was read. The Sept 20 release date is unverified: the dated sources read are 22 September or undated docs. No benchmark or user test of task success was found. The pruning approach differs from the default summary path, which is model-written and lossy by design. The plugin depends on Jev, a hosted early-access model. What an agent forgets is now a decision is the author's thesis.
Related work
- Note on Jev and Laya ↗Jev is the model compaction depends on.
Watch next
- Published task-success comparisons against summary compaction and Jev's availability and pricing.
Sources
- How it works (pi-fast-jev-compaction 0.2.0)cdn.jsdelivr.net
- Jev compaction (TypeSafe)jevtypesafeai.com
- Fast Jev compaction replaces summaries with verbatim context (ClaudeWorkshop)claudeworkshop.com
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 20 September 2026 at 11:08 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →