What it does
The harness, not Jev, runs commands through its existing permissions. In the author’s three-task subset, estimated cost across all three rounds, including learning, was 11.04% above three times the measured baseline round.
Agent tooling
JevDo
Reuse saved project actions for routine work, while Jev decides when a coding task needs the main model.
Bookmark: JevDo
Keep this in your collection.
Leave a noteWhat would you try with this?
: JevDo
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
Ask a developer to save and check a repeated project task, such as running tests, as an Action. On a later request, Jev can choose that saved action instead of asking the main AI to plan it again. The project’s tools run the commands and check the result.
Keep the main AI available for new code, explanations and failed tasks. This early plugin targets DeepSeek Harness and needs service access keys for live AI calls. Compare costs including the first learning round, not just later reuse. The author’s small experiment does not establish savings for your project.