Question 1

What is Jev eSketcher?

Accepted Answer

A generative painting instrument. You select part of a sketch, Jev decides which paint material it should get, and the material leaves the stream at the bottom of the screen, flies across the desk and paints itself in.

Question 2

How does Jev make decisions inside eSketcher?

Accepted Answer

Each question goes through one parallel-sampler query isolated against the same state. Jev returns a probability distribution over the declared options, the choice is the most likely option, and confidence is (n times p_max minus 1) over (n minus 1).

Question 3

How many sketches and paint materials ship with it?

Accepted Answer

105 procedural line-art sketches across 21 categories, each with named paintable regions, and 121 procedural paint materials across 13 behaviours, each with its own reveal animation such as watercolor blooms, chrome sweeps or glitter orbits.

Question 4

What is the tech stack?

Accepted Answer

FastAPI plus React and TypeScript. The browser holds the desk canvas, paint engine, Jev controller, Play and Sampling, while the backend owns Jev routes, JevClient, the catalog and SQLite, and alone calls TypeSafe Jev or the offline mock.

Question 5

Do I need a TypeSafe API key to run it?

Accepted Answer

Not necessarily: the backend can call TypeSafe Jev or an offline mock. The README excerpt in this digest is truncated before any install commands, so check the repository for current setup instructions.

Question 6

What is RLCD and why does it matter here?

Accepted Answer

RLCD stands for Reinforcement Learning for Calibrated Decisions, the approach TypeSafe documents for training Jev's probabilities against real outcomes so they are calibrated. TypeSafe does not publish the architecture, reward, loss or weights.

Question 7

Who built it and what are the limits of the documentation?

Accepted Answer

faizullah9181 published the MIT-licensed repo in September 2026 and last pushed it on 2026-09-27. The README is explicit that TypeSafe publishes Jev's behaviour only, so internals beyond the probability and confidence rules are not documented.

Jev eSketcher

Highlights

Questions about this entry

From the Jev guides

From the Jev AI model community