Jev in production › Agent context and memory

fast-jev-compaction-opencode

OpenCode V2 plugin asks Jev which completed tool calls and results are still needed, compacting agent context without a model-generated summary.

Open on GitHub ↗

The source publishes no measured number.
Use
Agent context and memory
Industry
Developer tools
Form
Open-source tool
Stage
Beta
Plugs into
OpenCode
Listed
2026-09-30
Found via
github
Repository
overbit/fast-jev-compaction-opencode
Stars
13
Forks
2
Last push
2026-10-03
Language
TypeScript
License
MIT

The README opens with

Instead of asking a generative model to rewrite old conversation history into a summary, this plugin asks a decision model which completed tool calls and outputs are still useful.

This project ports the core approach from tamaratran/fast-jev-compaction to the OpenCode V2 plugin API, with hosted JEV and a working local LM Studio backend.

Status: experimental. The compaction strategy is intentionally conservative and falls back to OpenCode's native compaction when classification fails or the reduction is too small to justify an override. Supported today: hosted TypeSafe JEV and the LM Studio-compatible Jev-Style 2B v1 classifier. Jev-Style 0.8B v3 is documented below, but direct use of its scoring runtime is not yet implemented by this plugin.

Badge

For the project's own README, linking back here:

Listed in Jev in production

Also used for agent context and memory