Tool Prune
Select relevant agent tools before showing their schemas to an LLM.
What Jev decides
Which tool description best matches the request.
Tool Prune ranks tool descriptions for a request, narrows the schemas sent to an agent, and can dispatch a confident match directly.
- Jev role
- Rank available tools against the current request when the TypeSafe engine is selected.
- Input
- The user request and the supplied tool names and descriptions.
- Output
- A selected tool and scores used for filtering or direct dispatch.
- Decision frequency
- Before an agent receives candidate tool schemas or dispatches a request.
- Fallback
- Use the offline TurboQuant engine when the Jev API is not configured.
How it works
Overview
Tool Prune offers an offline TurboQuant engine by default and an optional TypeSafe Jev engine when an API key is present. It can return one matching tool, filter a larger tool set, or send a confident request to a handler without an LLM turn. The repository reports benchmark results for both engines; the Jev path is optional rather than required for every use.
- Language
- JavaScript / Python
- License
- MIT
- Author
- hemanth
Sources
Related projects
jev-ultrafast
A browser agent where Jev selects an operation and a compatible page element from a live indexed action space.
What Jev decides
Which supported browser operation to run, and which compatible visible element to target
Jev Review
A code-review pipeline assembled from small typed judgments instead of one long reviewer prompt.
What Jev decides
What is risky, which files and evidence matter, what could fail, how severe it is, and who should review it.
Distill
Route coding-agent work, reasoning effort, and context retention through bounded Jev decisions.
What Jev decides
Which configured model and effort should handle a call, what evidence deserves attention, whether another attempt is useful, and what context should be retained.