Video unavailable. Open the source below.
NewDeerFlow
An agent harness with opt-in Jev tools for batched text classification and decision support.
Added to Jevfast
How it uses Jev
An opt-in DeerFlow extension exposes a classify_texts tool. Its Jev backend batches texts into state and asks one Choice question per positional item. The tool validates returned labels and maps them back to caller IDs; the agent writes labels to files itself.
What Jev decides
Source-defined request (illustrative)Example answers · not a recorded Jev response · Source ↗
Question 1 · item 1
YOUR APP
INSTRUCTION
Classify only the text with positional id 1 using the supplied category descriptions.
STATE
Representative README support-ticket batch with categories billing and technical. The first text is “My card was charged twice”. Wire state uses positional id 1 and question item_1; caller row IDs remain outside the wire identifiers.
JEV · CHOICE
- billing
- technical
App workflow
- Prepare items
The agent reads a file and supplies stable row IDs, text and category descriptions to the opt-in tool.
- Classify bounded batches
The Jev backend creates one Choice per positional item and returns validated labels and statuses.
- Apply the labels
The tool maps results back to caller IDs; the agent decides how to write them to the source file.
Why it is interesting
Batch classification separates fast typed judgments from file manipulation. Caller IDs never become wire question keys, and the tool reports per-item errors instead of treating a partial response as a complete result.