Question 1

What is el-jev and what does it do?

Accepted Answer

A fast typed-decision sidecar and a GitHub Copilot CLI pre-turn hook that mimics TypeSafe Jev's decision primitives — choice, noul and score — on Azure AI Foundry Cohere with keyless Entra ID auth, designed for auditable abstention first.

Question 2

Is el-jev the same thing as TypeSafe Jev?

Accepted Answer

No. The README's comparison is blunt: TypeSafe Jev is a purpose-built System One model whose decisions application code branches on, while el-jev recreates that bounded-decision pattern using Cohere relevance scores, and its Copilot hook only adds pre-turn guidance to the prompt.

Question 3

Does the Copilot hook control what Copilot does?

Accepted Answer

No. Copilot's LLM still runs on every turn and remains responsible for choosing tools and actions; the hook adds guidance that Copilot may follow or override. The decision effect depends on the harness, not on el-jev.

Question 4

Where can el-jev run?

Accepted Answer

The supported deployment is a local sidecar: the daemon binds to 127.0.0.1:8787 and the hook, CLI, MCP server or another local application can call it. A shared remote service is architecturally possible but is not supported as-is.

Question 5

Does it need an API key for the model?

Accepted Answer

The project description says keyless Entra ID auth against Azure AI Foundry Cohere, so the decision call is not keyed by a plaintext provider key. v0.2.0 also hardened the loopback daemon with Origin, Host and Content-Type checks.

Question 6

What changed in v0.2.0?

Accepted Answer

Unified runtime config resolution (environment over config.json over defaults) with repo-local state in .eljev/, configure and hook lifecycle commands, built-in Cohere-backed choice, noul and score decisions with fail-closed handling, and one shared apply_gate for all shapes.

Question 7

How fast is it in practice?

Accepted Answer

The README reports warm Cohere rerank p50 around 175-183 ms on common flows, measured on a single machine on 2026-09-24 and 2026-09-30. Those are not cross-environment SLAs, and the digest truncates the rest of the latency table.

El Jev

Highlights

Questions about this entry

From the Jev guides

From the Jev AI model community