GoKawiilDeveloper builds 'jevchat', a character-by-character chatbot powered by an LLM named Jev
A hobby project called jevchat turns a model referred to as Jev into a chat interface by asking it, one symbol at a time, to predict the next character of its reply given the conversation so far. The tool samples from a probability distribution over a chosen alphabet (letters, truncated tokens, etc.) using configurable sampling strategies, displaying live generation stats and the top candidate symbols as it works. The creator built it with Claude's help by describing the sampling algorithms and having the AI implement them.
GoKawiil's interpretation of the reporting above, not reported fact.
The project is a playful but instructive demonstration of how token-by-token (or symbol-by-symbol) generation actually works under the hood, making the mechanics of sampling and probability distributions visible rather than hidden. It's also a small case study in AI-assisted coding, where a human designs the concept and algorithm and an AI model handles implementation.
- jevchat generates text one symbol at a time by repeatedly querying an LLM called Jev for the next-character probability distribution.
- The tool supports multiple alphabets and sampling strategies, letting users experiment with how output quality and speed change.
- The entire implementation was built via an AI-assisted coding session, with Claude writing code based on the author's algorithmic descriptions.
Source: github.com, 2026-09-20
Published there as: “I turned Jev into a (lousy) chatbot”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.