OPEN-SOURCE STUDIO · AI-ASSISTED WORKFLOWS

Build AI workflows
with clear decisions.

Jev Flow is an open-source studio for AI-assisted workflows. Jev answers a focused question with a structured result, such as a category, probability, or score. The workflow checks the answer, applies your rules, and selects the next step in code. An execution trace shows each step.

Watch the walkthrough ↘Run it locally →

Run the application locally. Connect a Jev-compatible backend for live typed judgments.

Jev Flow Studio showing an editable workflow, input, and execution trace — Edit a workflow, provide input, and inspect its route in the Studio.

01 / THE PROBLEM

A prompt alone hides how a decision was made.

A single prompt can classify input, choose a path, and draft an action at once. Jev Flow separates those operations into steps you can inspect and test.

Jev provides a typed judgment. Rules in code choose the route.

External effects require explicit gates.

02 / HOW IT WORKS

From input to outcome, one step at a time.

Jev provides typed answers where a flow uses them. Code validates each answer, applies your rules, and records the route.

{ }

Accept the input

Define the fields a workflow accepts and validate the incoming data against that schema.

INPUT

Get a typed judgment

Ask Jev a bounded question. It returns a category, probability, or score in a known shape.

TYPED JUDGMENT

◇

Validate and route

Code checks the type and range, applies deterministic rules, and selects the next branch.

CODE ROUTE

↗

Inspect the trace

Follow the executed path and see answers, timing, and usage when available.

EXECUTION TRACE

The boundary: Jev interprets a limited question; the application decides what to do with its answer.

SEE JEV FLOW IN ACTION

Follow the flow from input to outcome.

See how typed judgments meet code-owned rules, then tour the Studio, Compendium, a live Arena comparison, games, and the simulated road. Each version has two-speaker narration and captions burned into the video.

Narration languageWebVTT captions ↗Exact transcript ↗

03 / COMPARE

One input. Two kinds of answers.

Send the same text to Jev and a selected language model. Jev returns structured answers to specific questions; the model replies in free text. Compare their answers and response times. Token use appears when available, and cost is labeled as an estimate.

Jev Battle Arena comparison workspace with a prompt form and two separate answer panels — **Arena:** Choose the engines, enter a prompt, and select Compare to run them.

2answer panes

Each answer stays beside its source and measured response time. Token use appears when available; cost is shown as an estimate.

Run the local application ↗

04 / EXPLORE

Find a flow, then try it.

In the Compendium, filter configurations by pattern and use case, open their JSON, test them in the Studio, and install the ones you want to edit.

388,080configurations passed validity and unique-ID checks

The Compendium generates configurations on demand. Its certificate records validity and unique IDs for all 388,080. Choose one as a starting point, then adapt it in the Studio.

Inspect the certificate ↗

Compendium catalog with filters and flow configurations generated on demand — Filter, open, test, or install a configuration in the local application.

Two-track self-driving Cart simulation with decision overlays — CART / SIMULATIONTest decisions on a changing road.Change traffic and obstacles, then inspect lane, speed, and safety decisions in the driving simulation.

Blast Garden in Jev Labs with game state and decision controls — LABS / BLAST GARDENTry bounded decisions in a game.Inspect each proposed action against the game state.

Chess Lab board with a selected pawn and highlighted legal moves — LABS / CHESSPlay within legal moves.The local chess engine validates every move.

MORE TO EXPLORE

Tools for decisions and testing.

Labs includes spam screening, evidence checks, rescue scenarios, chess, and Blast Garden. Other tools help you build, test, and connect workflows.

SHIP PACK / 10 + 11

Reusable judgments.

Ship Pack provides 10 Jevlet definitions and 11 named gates for common decision tasks.

Inspect source ↗

SUITE / 12 + 52

Test fixed cases.

Run 52 repeatable cases across 12 scenarios and compare each answer with its fixed expected result.

Inspect source ↗

GAMES / PROMPT TOOLS

Explore games and context.

Play tic-tac-toe against local minimax, try fictional combat with optional Jev judgments, or advance a city one tick at a time with local rules or a live Jev call. Prompt tools use 16 patterns and help judge cache reuse and content priority.

Games source ↗Prompt tools ↗

STUDIO / CONNECTIONS

Connect workflow steps.

Use webhooks, subflows, loops, error routes, and design chat in the Studio. Code gates external actions.

Engine source ↗Chat source ↗

Connect a Jev-compatible backend for typed judgments, or set up optional local Laya. Configure an LLM separately for Arena and design chat.

05 / GET STARTED

Run Jev Flow locally.

Install Node.js 20 or later, clone the repository, and start the app. Explore the Studio, examples, and local tests without credentials. Connect a Jev-compatible backend for typed judgments, or set up optional local Laya. Configure an LLM separately for Arena and design chat.

This website is a static guide; the Node application runs on your machine or protected server.

Full quickstart in README ↗Security boundary ↗

LOCAL SETUP

git clone https://github.com/daltonrpj/jev-flow.git
cd jev-flow
npm ci
npm start

Open http://127.0.0.1:8723/jev/flows in your browser.

EDITABLE EXAMPLE FLOWSSupport triage ↗Spam screening ↗Refund intake ↗Anomaly review ↗Interactive labs ↗

Build AI workflowswith clear decisions.