CreativeJeved
New
Jeved
Natural-language sensors and configurable rules for SillyTavern replies.
Added to Jevfast
How it uses Jev
Jeved measures configured aspects of SillyTavern user messages and narrator replies. For the built-in `speaks` sensor, it sends `player_message` and `latest_turn` as state and asks Jev for a Score from 0 to 4 on how much the narrator writes for the user. The extension stores the returned score, then evaluates rules over measured sensor values to trigger actions such as adding a directive or running a script.
What Jev decides
Built-in Speaks for you sensor exampleExample answers · not a recorded Jev response · Source ↗
Question 1 · speaks
YOUR APP
INSTRUCTION
How much does the narrator reply write what only the user does, says, or thinks?
STATE
The built-in sensor reads one player message and one narrator reply, with no extra context group. The state fields are `player_message` and `latest_turn`; the sensor asks how much the reply writes what only the user does, says, or thinks.
JEV · SCORE
0
App workflow
- Build sensor requests
The extension groups needed sensors that share message windows and context settings. It builds state from the selected player and narrator messages and sends the configured sensor questions to the selected decision provider.
- Save measurements
Jev's returned sensor scores are written to the chat message and the chat is saved.
- Evaluate rules
Before generation, Jeved evaluates enabled rules against the stored measurements and can apply the matching actions, such as list changes, scripts, or reply directives.
Why it is interesting
The extension turns natural-language checks on chat turns into reusable sensor values. Those values feed configurable rules and actions, so a user can automate narrator behavior without writing a separate classifier for each condition.