A useful ninth category is browser-grounded research, where the decision is which typed retrieval operation to run next—not what prose to generate. We built Jev Social around that boundary: Jev selects a read-only Instagram/TikTok/LinkedIn action, socai executes it in the user’s Chrome, and a slower model synthesizes only captured evidence. The evaluation unit is route accuracy + evidence coverage + end-to-end completion, with partial runs preserved.
Save these JEV resources. Thank me later.
- Demos & use cases: madewithjev.com
- What is JEV: madewithjev.com/what-is-jev
- What is JEV Engineering: https://madewithjev.com/what-is-jev-engineering
- Tools: madewithjev.com/tools
- GitHub repos: https://madewithjev.com/github-repos - JEV Agent Skills: https://madewithjev.com/skills - JEV Agents & browser use: https://madewithjev.com/categories/agents-and-browsers
- Awesome-JEV repo: github.com/kraayenjon/awesome-jev
-JEV Resources & Tutorials: https://madewithjev.com/resources
Fast Jev Compaction is the one I'd try first. Scoring tool results and keeping the good ones verbatim beats summarizing them into mush.
A useful ninth category is browser-grounded research, where the decision is which typed retrieval operation to run next—not what prose to generate. We built Jev Social around that boundary: Jev selects a read-only Instagram/TikTok/LinkedIn action, socai executes it in the user’s Chrome, and a slower model synthesizes only captured evidence. The evaluation unit is route accuracy + evidence coverage + end-to-end completion, with partial runs preserved.