DSPy introduces ReAnchor, an optimizer that runs programs on TypeSafe's System One decision model Jev

dbreunig · x · 2026-09-25

Drew Breunig and Isaac Miller released ReAnchor, a DSPy optimizer built for TypeSafe's Jev and other System One models.

What Jev is: Not an LLM — TypeSafe calls it a "System One" frontier model, a decision engine returning true/false, multiple-choice, or ranked answers that software can use directly, extremely cheap and fast.

Why it fits DSPy: DSPy separates task definition from implementation, so switching models means changing config and re-optimizing. Many already use DSPy to turn LLMs into callable functions optimized via GEPA or MIPROv2 — Jev's fast structured decisions map directly onto this pattern.

Usage: Install with pip install -U "dspy[typesafe]" and define/optimize Jev-powered programs without changing existing task definitions. TypeSafe's Diogo Almeida hinted more DSPy-aligned work is coming.

Related event: DSPy 3.4.0 Adds Native Jev Support and ReAnchor Optimizer(2 posts)→

Original post →

More from coding & agent

coding & agent channel →