NewOh My Pi
Coding-agent fork with a native TypeSafe Jev judgment provider used for bounded agent decisions.
Added to Jevfast
NewHow it uses Jev
Oh My Pi has a reusable typed judgment interface with a native TypeSafe backend that forwards state and questions to System One. Its automatic thinking feature can ask Jev to classify each coding-agent request into a reasoning-effort tier; the resolver also supports judgments for unexpected stops and AI-assisted Git staging.
What Jev decides
Classify coding request difficulty for automatic thinking effortExample answers · not a recorded Jev response · Source ↗
Question 1 · level
YOUR APP
INSTRUCTION
Choose reasoning effort by inherent task difficulty, not the request's wording or politeness.
STATE
The state is the user's request text after preprocessing. The question is a fixed Choice; the `max` option is included only when the target model supports that effort and the configured ceiling allows it.
JEV · CHOICE
- low
- medium
- high
- xhigh
- max (conditional)
App workflow
- Enable automatic effort
When the coding agent's thinking level is `auto`, the classifier preprocesses the current user request and asks the configured judgment backend for a difficulty tier.
- Map the judgment to model effort
The selected tier is clamped to the active model's supported effort range. If judgment fails, the caller falls back to a concrete effort and continues the turn.
- Reuse typed judgments elsewhere
The same judgment interface also supports user-triggered AI Git staging, where file and hunk relevance are assessed through separate typed questions.
Why it is interesting
The same small judgment protocol is reused for agent control decisions. In AI staging, Jev can first select relevant changed files and then assess their individual hunks, while automatic thinking maps request difficulty to a concrete model effort and falls back if classification fails.