Jev in production › Model and agent routing

Anchor (QpiAI)

Extracts facts and a confidence score from policy requests with Jev, falling back to GPT for about a third of cases while the rest finish in 350ms (Anchor).

Open on GitHub ↗

350msmeasured, as published by the source
Use
Model and agent routing
Industry
AI infrastructure
Form
Open-source tool
Stage
In production
Listed
2026-09-25
Found via
github
Repository
qpiai/anchor
Stars
10
Forks
2
Last push
2026-09-24
Language
Python
License
MIT

The README opens with

Anchor turns a policy document into a formal policy, then checks a question and answer against it.

GPT writes the policy (variables and rules). Jev, TypeSafe's fast system-one model, extracts the facts and a confidence score. Z3 is the deterministic decider. When Jev is unsure of a fact the decision needs, or of any fact behind an approval, GPT reads the request and its reading is used instead. About a third of requests take that path; the rest finish in about 350 ms. Facts a trusted system supplies override the text.

Badge

For the project's own README, linking back here:

Listed in Jev in production

Also used for model and agent routing