Skip to content

mesa-anyjev

mesa-anyjev turns every choice on the way from a dataset to an ontology-grounded AVU into a typed question with a probability and an explicit calibration level, using AnyJev over an open model. A reasoning model only plans (which ontologies are in play, what to search); deterministic code drives mesa-mcp's OBO/OLS tools; AnyJev answers the multiple-choice steps; every decision is recorded in a provenance sidecar next to the AVU history kept by mesa-ducklake. Hosted Jev through MotherDuck's prompt_jev is a third, policy-gated provider for the same questions.

It is developed by idss-mesa at the University of New Mexico and released under the MIT license. Status: pre-alpha (milestone M0).

Documentation

This site is an Open Knowledge Format (OKF) v0.2 knowledge bundle; the corpus is available to agents at llms.txt and llms-full.txt.

  • Getting started - Install mesa-anyjev, reach the CARC gateway and configure it.
  • Concepts - The decision graph, calibration levels and the write policy, provenance, hosted providers.
  • Bench - Measured numbers per question and level, every cell naming its results JSON.
  • Develop - Architecture and the hermetic test suite.
  • About - How agents should consume this site, licenses, and the change log.