Jev by BlackEyeTech

BlackEyeTech · first product · early-stage

Choose the right AI model for the task in front of you.

Jev is a VS Code extension that reads the coding task you describe and the small slice of project context around it, then recommends an appropriate AI model. It shows a fit score and the findings behind it, so the decision is yours to make with reasons, not a black box.

Jev is an advisor. It does not switch models for you, edit your files, or run coding tasks on its own.

See how it works View repository
Runs in
VS Code
Provides
Fit scores & findings
Control
Stays with you
Jev Illustrative demo

Task

Review the authentication change for architecture and security issues before I extend it.

Model fit

  • Claude 91
  • GPT 84
  • Gemini 79

Finding

Architecture Confidence: moderate

Password policy and persistence live in the API controller rather than a service layer.

Evidence: src/api/authController.ts

Illustrative demo built for this page. Not a real screenshot. Scores shown are Jev fit scores for the task, not provider benchmarks.

How it works

Four steps, and you keep the wheel

Jev sits beside your editor and answers one question well: which model suits this task, and why.

  1. Describe the task

    Type what you are about to do in plain language, right in the Jev sidebar.

  2. Collect a small context

    Jev gathers only what is relevant: the current file, your selection, and a light project structure. It keeps that context minimal.

  3. Analyze and recommend

    Jev evaluates the task and produces a ranked model recommendation with fit scores, using a supported provider you have configured.

  4. Review the reasoning

    Open a finding to see its evidence, the recommendation, and a short explanation of why. Accept, dismiss, or ask for more.

Jev recommends and explains. It never changes your model, your files, or your code on its own.

Features

What Jev does

A focused set of capabilities for the decision layer around AI-assisted coding.

  • Model fit scores

    A ranked recommendation for your task. Scores express how well a model fits the task, not a provider benchmark.

  • Explainable findings

    Every significant finding carries its evidence, a recommendation, and a plain reason. Confidence is a product signal, not a probability of correctness.

  • Minimal context

    Jev sends as little as it can: the current file, your selection, and a light project tree. Keys stay on the server, never in the extension.

  • Non-blocking findings

    Findings surface in a quiet notification with Review, Why, and Dismiss. Repeated findings are throttled, so Jev does not spam you.

  • Provider integration

    Jev can integrate with supported model providers that you configure. Not every provider or model is available, and only the configured one is invoked.

  • You stay in control

    The developer decides. Jev never takes an action on your behalf without your explicit say-so.

Example workflow

One task, start to finish

This walkthrough follows the sample project that ships with the repository.

  1. Open a project in VS Code and open Jev from the activity bar.
  2. Focus the file you are about to change, for example src/api/authController.ts.
  3. Enter the task in plain language:
Review the authentication change for architecture and security issues before I extend it.
  1. Read the model recommendation and the finding, then open Review and Why to see the evidence.
  2. Decide. Accept the recommendation, dismiss the finding, or ask for more explanation.

What a finding looks like (illustrative demo)

Business rules and persistence in the controller

Architecture Affected: src/api/authController.ts

Why: password policy and storage sit in the API controller while a service layer exists but goes unused, which couples transport to business rules.

The demo project is intentionally inconsistent so a finding has something real to describe. Wording from a live model varies, and the exact result depends on the provider you have configured.

Installation

Try it from source today

Jev is early-stage. There is no public release yet: the VS Code Marketplace listing and a download are coming soon. You can run it from the repository in the meantime.

Run it locally

  1. Clone the repository and install dependencies.
  2. Create a .env from the example and set your provider key.
  3. Start the local server, then launch the extension host.
git clone https://github.com/Dawit749/jev.git
cd jev
npm install
cp .env.example .env
npm run dev:server

Requires Node.js 20 or newer. The server binds to the loopback address only. Provider keys stay in the server environment and are never handed to the extension.

Limitations

What Jev is not

Honest limits, stated up front. Jev is a small, early-stage advisor, and it says so.

  • An advisor, not an autonomous coder. Jev does not write or edit your files, and does not run coding tasks for you.
  • No automatic model switching. Jev does not change the model in your editor's selector or touch editor internals.
  • Findings are recommendations. They are guidance grounded in the files they cite, not guaranteed truths. Review them before you act.
  • Only the configured provider is used. Other models can appear as fit comparisons and are marked as not connected. Not every provider or model is available.
  • Early-stage software. Interfaces and behavior may change, and live end-to-end verification through the extension host is still in progress.
  • No invented proof. There are no customers, downloads, benchmarks, or certifications to show yet, and Jev does not claim any.

Contact & feedback

Tell us where Jev gets it wrong

Jev is in early development, and feedback shapes it. Bug reports, confusing findings, and missing use cases are all useful.

Open an issue Browse the source